Download Star Formation in Galaxies Along the Hubble Sequence

STAR FORMATION IN GALAXIES ALONG THE HUBBLE SEQUENCE
Robert C. Kennicutt, Jr
Steward Observatory, The University of Arizona
Tucson, Arizona 85721
[email protected]
arXiv:astro-ph/9807187v1 17 Jul 1998
ABSTRACT
Observations of star formation rates (SFRs) in galaxies provide vital clues to the physical nature
of the Hubble sequence, and are key probes of the evolutionary histories of galaxies. The focus of
this review is on the broad patterns in the star formation properties of galaxies along the Hubble
sequence, and their implications for understanding galaxy evolution and the physical processes
that drive the evolution. Star formation in the disks and nuclear regions of galaxies are reviewed
separately, then discussed within a common interpretive framework. The diagnostic methods used
to measure SFRs are also reviewed, and a self-consistent set of SFR calibrations is presented as a
aid to workers in the field.
KEY WORDS: galaxy evolution, starbursts, spiral galaxies, star formation rates, stellar populations
To appear in Vol. 36 of Annual Review of Astronomy and Astrophysics
1
INTRODUCTION
One of the most recognizable features of galaxies along the Hubble sequence is the wide range in
young stellar content and star formation activity. This variation in stellar content is part of the
basis of the Hubble classification itself (Hubble 1926), and understanding its physical nature and
origins is fundamental to understanding galaxy evolution in its broader context. This review deals
with the global star formation properties of galaxies, the systematics of those properties along the
Hubble sequence, and their implications for galactic evolution. I interpret “Hubble sequence” in
this context very loosely, to encompass not only morphological type but other properties such as
gas content, mass, bar structure, and dynamical environment, which can strongly influence the
large-scale star formation rate (SFR).
Systematic investigations of the young stellar content of galaxies trace back to the early studies
of resolved stellar populations by Hubble and Baade, and analyses of galaxy colors and spectra
by Stebbins, Whitford, Holmberg, Humason, Mayall, Sandage, Morgan, and de Vaucouleurs (see
Whitford 1975 for a summary of the early work in this field). This piecemeal information was
synthesized by Roberts (1963), in an article for the first volume of the Annual Review of Astronomy
and Astrophysics. Despite the limited information that was available on the SFRs and gas contents
1
of galaxies, Roberts’ analysis established the basic elements of the contemporary picture of the
Hubble sequence as a monotonic sequence in present-day SFRs and past star formation histories.
Quantifying this picture required the development of more precise diagnostics of global SFRs in
galaxies. The first quantitative SFRs were derived from evolutionary synthesis models of galaxy
colors (Tinsley 1968, 1972, Searle et al 1973). These studies confirmed the trends in SFRs and star
formation histories along the Hubble sequence, and led to the first predictions of the evolution of
the SFR with cosmic lookback time. Subsequent modelling of blue galaxies by Bagnuolo (1976),
Huchra (1977), and Larson & Tinsley (1978) revealed the importance of star formation bursts in
the evolution of low-mass galaxies and interacting systems. Over the next decade the field matured
fully, with the development of more precise direct SFR diagnostics, including integrated emissionline fluxes (Cohen 1976, Kennicutt 1983a), near-ultraviolet continuum fluxes (Donas & Deharveng
1984), and infrared continuum fluxes (Harper & Low 1973, Rieke & Lebofsky 1978, Telesco &
Harper 1980). These provided absolute SFRs for large samples of nearby galaxies, and these were
subsequently interpreted in terms of the evolutionary properties of galaxies by Kennicutt (1983a),
Gallagher et al (1984), and Sandage (1986).
Activity in this field has grown enormously in the past decade, stimulated in large part by two
major revelations. The first was the discovery of a large population of ultraluminous infrared
starburst galaxies by the Infrared Astronomical Satellite (IRAS) in the mid-1980’s. Starbursts had
been identified (and coined) from groundbased studies (Rieke & Lebofsky 1979; Weedman et al
1981), but IRAS revealed the ubiquity of the phenomenon and the extreme nature of the most
luminous objects. The latest surge of interest in the field has been stimulated by the detection of
star forming galaxies at high redshift, now exceeding z = 3 (Steidel et al 1996, Ellis 1997). This
makes it possible to apply the locally calibrated SFR diagnostics to distant galaxies, and directly
trace the evolution of the SFR density and the Hubble sequence with cosmological lookback time.
The focus of this review is on the broad patterns in the star formation properties of galaxies,
and their implications for the evolutionary properties of the Hubble sequence. It begins with a
summary of the diagnostic methods used to measure SFRs in galaxies, followed by a summary of
the systematics of SFRs along the Hubble sequence, and the interpretation of those trends in terms
of galaxy evolution. It concludes with a brief discussion of the physical regulation of the SFR in
galaxies and future prospects in this field. Galaxies exhibit a huge dynamic range in SFRs, over
six orders of magnitude even when normalized per unit area and galaxy mass, and the continuity
of physical properties over this entire spectrum of activities is a central theme of this review.
With this broad approach in mind, I cannot begin to review the hundreds of important papers on
the star formation properties of individual galaxies, or the rich theoretical literature on this subject.
Fortunately, there are several previous reviews in this series that provide thorough discussions of
key aspects of this field. A broad review of the physical properties of galaxies along the Hubble
sequence can be found in Roberts & Haynes (1994). The star formation and evolutionary properties
of irregular galaxies are reviewed by Gallagher & Hunter (1984). The properties of IR-luminous
starbursts are the subject of several reviews, most recently those by Soifer et al (1987), Telesco
(1988), and Sanders & Mirabel (1996). Finally an excellent review of faint blue galaxies by Ellis
(1997) describes many applications to high-redshift objects.
2
2
DIAGNOSTIC METHODS
Individual young stars are unresolved in all but the closest galaxies, even with the Hubble Space
Telescope (HST), so most information on the star formation properties of galaxies comes from
integrated light measurements in the ultraviolet (UV), far-infrared (FIR), or nebular recombination
lines. These direct tracers of the young stellar population have largely supplanted earlier SFR
measures based on synthesis modelling of broadband colors, though the latter are still applied to
multicolor observations of faint galaxies. This section begins with a brief discussion of synthesis
models, which form the basis of all of the methods, followed by more detailed discussions of the
direct SFR tracers.
2.1
Integrated Colors and Spectra, Synthesis Modelling
The basic trends in galaxy spectra with Hubble type are illustrated in Figure 1, which shows
examples of integrated spectra for E, Sa, Sc, and Magellanic irregular galaxies (Kennicutt 1992b).
When progressing along this sequence, several changes in the spectrum are apparent: a broad rise
in the blue continuum, a gradual change in the composite stellar absorption spectrum from K-giant
dominated to A-star dominated, and a dramatic increase in the strengths of the nebular emission
lines, especially Hα.
Although the integrated spectra contain contributions from the full range of stellar spectral types
and luminosities, it is easy to show that the dominant contributors at visible wavelengths are
intermediate-type main sequence stars (A to early F) and G-K giants. As a result, the integrated
colors and spectra of normal galaxies fall on a relatively tight sequence, with the spectrum of any
given object dictated by the ratio of early to late-type stars, or alternatively by the ratio of young
(< 1 Gyr) to old (3–15 Gyr) stars. This makes it possible to use the observed colors to estimate
the fraction of young stars and the mean SFR over the past 108 –109 years.
The simplest application of this method would assume a linear scaling between the SFR and the
continuum luminosity integrated over a fixed bandpass in the blue or near-ultraviolet. Although
this may be a valid approximation in starburst galaxies, where young stars dominate the integrated
light across the visible spectrum, the approximation breaks down in most normal galaxies, where
a considerable fraction of the continuum is produced by old stars, even in the blue (Figure 1).
However the scaling of the SFR to continuum luminosity is a smooth function of the color of the
population, and this can be calibrated using an evolutionary synthesis model.
Synthesis models are used in all of the methods described here, so it is useful to summarize the
main steps in the construction of a model. A grid of stellar evolution tracks is used to derive
the effective temperatures and bolometric luminosities for various stellar masses as a function of
time, and these are converted into broadband luminosities (or spectra) using stellar atmosphere
models or spectral libraries. The individual stellar templates are then summed together, weighted
by an initial mass function (IMF), to synthesize the luminosities, colors, or spectra of single-age
populations as functions of age. These isochrones can then be added in linear combination to
synthesize the spectrum or colors of a galaxy with an arbitrary star formation history, usually
3
Figure 1: Integrated spectra of elliptical, spiral, and irregular galaxies, from Kennicutt (1992b).
The fluxes have been normalized to unity at 5500 Å.
parametrized as an exponential function of time. Although a single model contains at least four
free parameters, the star formation history, galaxy age, metal abundance, and IMF, the colors of
normal galaxies are well represented by a one-parameter sequence with fixed age, composition and
IMF, varying only in the time dependence of the SFR (Searle et al 1973, Larson & Tinsley 1978;
Charlot & Bruzual 1991).
Synthesis models have been published by several authors, and are often available in digital form.
An extensive library of models has been compiled by Leitherer et al (1996a), and the models are
described in a companion conference volume (Leitherer et al 1996b). Widely used models for star
forming galaxies include those of Bruzual & Charlot (1993), Bertelli et al (1994), and Fioc & RoccaVolmerange (1997). Leitherer & Heckman (1995) have published an extensive grid of models that
is optimized for applications to starburst galaxies.
The synthesis models provide relations between the SFR per unit mass or luminosity and the
integrated color of the population. An example is given in Figure 2, which plots the SFR per
unit U , B, and V luminosity as functions of U − V color, based on the models of Kennicutt et al
4
Figure 2: Relationship between SFR per unit broadband luminosity in the U BV passbands and
integrated color, from the evolutionary synthesis models of Kennicutt et al (1994). The models are
for 10-billion-year-old disks, a Salpeter IMF, and exponential star formation histories. The U , B,
and V luminosities are normalized to those of the Sun in the respective bandpasses.
(1994). Figure 2 confirms that the broadband luminosity by itself is a poor tracer of the SFR; even
the SFR/LU ratio varies by more than an order of magnitude over the relevant range of galaxy
colors. However the integrated color provides a reasonable estimate of the SFR per unit luminosity,
especially for the bluer galaxies.
SFRs derived in this way are relatively imprecise, and are prone to systematic errors from reddening
or from an incorrect IMF, age, metallicity, of star formation history (Larson & Tinsley 1978).
Nevertheless, the method offers a useful means of comparing the average SFR properties of large
samples of galaxies, when absolute accuracy is not required. The method should be avoided in
applications where the dust content, abundances, or IMFs are likely to change systematically across
a population.
2.2
Ultraviolet Continuum
The limitations described above can be avoided if observations are made at wavelengths where the
integrated spectrum is dominated by young stars, so that the SFR scales linearly with luminosity.
The optimal wavelength range is 1250–2500 Å, longward of the Lyα forest but short enough to
minimize spectral contamination from older stellar populations. These wavelengths are inaccessible
from the ground for local galaxies (z < 0.5), but the region can be observed in the redshifted spectra
of galaxies at z ∼1–5. The recent detection of the redshifted UV continua of large numbers of z > 3
5
galaxies with the Keck telescope has demonstrated the enormous potential of this technique (Steidel
et al 1996).
The most complete UV studies of nearby galaxies are based on dedicated balloon, rocket, and
space experiments (Smith & Cornett 1982, Donas & Deharveng 1984, Donas et al 1987, 1995, Buat
1992, Deharveng et al 1994). The database of high-resolution UV imaging of galaxies is improving
rapidly, mainly from HST (Meurer et al 1995, Maoz 1996) and the Ultraviolet Imaging Telescope
(Smith et al 1996, Fanelli et al 1997). An atlas of UV spectra of galaxies from the International
Ultraviolet Explorer has been published by Kinney et al (1993). A recent conference volume by
Waller et al (1997) highlights recent UV observations of galaxies.
The conversion between the UV flux over a given wavelength interval and the SFR can be derived
using the synthesis models described earlier. Calibrations have been published by Buat et al (1989),
Deharveng et al (1994), Leitherer et al (1995b), Meurer et al (1995), Cowie et al (1997), and Madau
et al (1998), for wavelengths in the range 1500–2800 Å. The calibrations differ over a full range of
∼0.3 dex, when converted to a common reference wavelength and IMF, with most of the difference
reflecting the use of different stellar libraries or different assumptions about the star formation
timescale. For integrated measurements of galaxies, it is usually appropriate to assume that the
SFR has remained constant over timescales that are long compared to the lifetimes of the dominant
UV emitting population (<108 yr), in the “continuous star formation” approximation. Converting
the calibration of Madau et al (1998) to a Salpeter (1955) IMF with mass limits 0.1 and 100 M⊙
yields:
SFR (M⊙ yr −1 ) = 1.4 × 10−28 Lν (ergs s−1 Hz−1 ).
(1)
For this IMF, the composite UV spectrum happens to be nearly flat in Lν , over the wavelength
range 1500–2800 Å, and this allows us to express the conversion in Equation 1 in such simple
form. The corresponding conversion in terms of Lλ will scale as λ−2 . Equation 1 applies to
galaxies with continuous star formation over timescales of 108 years or longer; the SFR/Lν ratio
will be significantly lower in younger populations such as young starburst galaxies. For example,
continuous burst models for a 9 Myr old population yield SFRs that are 57% higher than those
given in Equation 1 (Leitherer et al 1995b). It is important when using this method to apply an
SFR calibration that is appropriate to the population of interest.
The main advantages of this technique are that it is directly tied to the photospheric emission of the
young stellar population, and it can be applied to star forming galaxies over a wide range of redshifts.
As a result, it is currently the most powerful probe of the cosmological evolution in the SFR
(Madau et al 1996, Ellis 1997). The chief drawbacks of the method are its sensitivity to extinction
and the form of the IMF. Typical extinction corrections in the integrated UV magnitudes are 0–3
magnitudes (Buat 1992, Buat & Xu 1996). The spatial distribution of the extinction is very patchy,
with the emergent UV emission being dominated by regions of relatively low obscuration (Calzetti
et al 1994), so calibrating the extinction correction is problematic. The best determinations are
based on two-component radiative transfer models which take into account the clumpy distribution
of dust, and make use of reddening information from the Balmer decrement or IR recombination
lines (e.g., Buat 1992, Calzetti et al 1994, Buat & Xu 1996, Calzetti 1997).
6
The other main limitation, which is shared by all of the direct methods, is the dependence of
the derived SFRs on the assumed form of the IMF. The integrated spectrum in the 1500–2500 Å
range is dominated by stars with masses above ∼5 M⊙ , so the SFR determination involves a
large extrapolation to lower stellar masses. Fortunately there is little evidence for large systematic
variations in the IMF among star forming galaxies (Scalo 1986, Gilmore et al 1998), with the
possible exception of IR-luminous starbursts, where the UV emission is of little use anyway.
2.3
Recombination Lines
Figure 1 shows that the most dramatic change in the integrated spectrum with galaxy type is a
rapid increase in the strengths of the nebular emission lines. The nebular lines effectively re-emit
the integrated stellar luminosity of galaxies shortward of the Lyman limit, so they provide a direct,
sensitive probe of the young massive stellar population. Most applications of this method have
been based on measurements of the Hα line, but other recombination lines including Hβ, Pα, Pβ,
Brα, and Brγ have been used as well.
The conversion factor between ionizing flux and the SFR is usually computed using an evolutionary
synthesis model. Only stars with masses >10 M⊙ and lifetimes <20 Myr contribute significantly
to the integrated ionizing flux, so the emission lines provide a nearly instantaneous measure of
the SFR, independent of the previous star formation history. Calibrations have been published
by numerous authors, including Kennicutt (1983a), Gallagher et al (1984), Kennicutt et al (1994),
Leitherer & Heckman (1995), and Madau et al (1998). For solar abundances and the same Salpeter
IMF (0.1–100 M⊙ ) as was used in deriving equation [1], the calibrations of Kennicutt et al (1994)
and Madau et al (1998) yield:
SFR (M⊙ yr −1 ) = 7.9 × 10−42 L(Hα) (ergs s−1 ) = 1.08 × 10−53 Q(H 0 ) (s−1 ).
(2)
where Q(H 0 ) is the ionizing photon luminosity, and the Hα calibration is computed for Case B
recombination at Te = 10000 K. The corresponding conversion factor for L(Brγ) is 8.2 × 10−40
in the same units, and it is straightforward to derive conversions for other recombination lines.
Equation 2 yields SFRs that are 7% lower than the widely used calibration of Kennicutt (1983a),
with the difference reflecting a combination of updated stellar models and a slightly different IMF
(Kennicutt et al 1994). As with other methods, there is a significant variation among published
calibrations (∼30%), with most of the dispersion reflecting differences in the stellar evolution and
atmosphere models.
Large Hα surveys of normal galaxies have been published by Cohen (1976), Kennicutt & Kent
(1983), Romanishin (1990), Gavazzi et al (1991), Ryder & Dopita (1994), Gallego et al (1995),
and Young et al (1996). Imaging surveys have been published by numerous other authors, with
some the largest including Hodge & Kennicutt (1983), Hunter & Gallagher (1985), Ryder & Dopita
(1993), Phillips (1993), Evans et al (1996), González Delgado et al (1997), and Feinstein (1997).
Gallego et al (1995) have observed a complete emission-line selected sample, in order to measure the
volume-averaged SFR in the local universe, and this work has been applied extensively to studies
of the evolution in the SFR density of the universe (Madau et al 1996).
7
The primary advantages of this method are its high sensitivity, and the direct coupling between
the nebular emission and the massive SFR. The star formation in nearby galaxies can be mapped
at high resolution even with small telescopes, and the Hα line can be detected in the redshifted
spectra of starburst galaxies to z≫2 (e.g. Bechtold et al 1997). The chief limitations of the method
are its sensitivity to uncertainties in extinction and the IMF, and to the assumption that all of
the massive star formation is traced by the ionized gas. The escape fraction of ionizing radiation
from individual HII regions has been measured both directly (Oey & Kennicutt 1997) and from
observations of the diffuse Hα emission in nearby galaxies (e.g., Hunter et al 1993, Walterbos
& Braun 1994, Kennicutt et al 1995, Ferguson et al 1996, Martin 1997), with fractions of 15–
50% derived in both sets of studies. Thus it is important when using this method to include the
diffuse Hα emission in the SFR measurement (Ferguson et al 1996). However the escape fraction
from a galaxy as a whole should be much lower. Leitherer et al (1995a) directly measured the
redshifted Lyman continuum region in four starburst galaxies, and they derived an upper limit of
3% on the escape fraction of ionizing photons. Much higher global escape fractions of 50–94%,
and local escape fractions as high as 99% have been estimated by Patel & Wilson (1995a, b),
based on a comparison of O-star densities and Hα luminosities in M33 and NGC 6822, but those
results are subject to large uncertainties, because the O-star properties and SFRs were derived from
U BV photometry, without spectroscopic identifications. If the direct limit of <3% from Leitherer
et al (1995a) is representative, then density bounding effects are a negligible source of error in
this method. However it is very important to test this conclusion by extending these types of
measurements to a more diverse sample of galaxies.
Extinction is probably the most important source of systematic error in Hα-derived SFRs. The
extinction can be measured by comparing Hα fluxes with those of IR recombination lines or the
thermal radio continuum. Kennicutt (1983a) and Niklas et al (1997) have used integrated Hα and
radio fluxes of galaxies to derive a mean extinction A(Hα) = 0.8–1.1 mag. Studies of large samples
of individual HII regions in nearby galaxies yield similar results, with mean A(Hα) = 0.5–1.8 mag
(e.g. Caplan & Deharveng 1986, Kaufman et al 1987, van der Hulst et al 1988, Caplan et al 1996).
Much higher extinction is encountered in localized regions, especially in the the dense HII regions
in circumnuclear starbursts, and there the near-IR Paschen or Brackett recombination lines are
required to reliably measure the SFR. Compilations of these data include Puxley et al (1990), Ho
et al (1990), Calzetti et al (1996), Goldader et al (1995, 1997), Engelbracht (1997), and references
therein. The Paschen and Brackett lines are typically 1–2 orders of magnitude weaker than Hα,
so most measurements to date have been restricted to high surface brightness nuclear HII regions,
but it is gradually becoming feasible to extend this approach to galaxies as a whole. The same
method can be applied to higher-order recombination lines or the thermal continuum emission at
submillimeter and radio wavelengths. Examples of such applications include H53α measurements
of M82 by Puxley et al (1989), and radio continuum measurements of disk galaxies and starbursts
by Israel & van der Hulst (1983), Klein & Grave (1986), Turner & Ho (1994), and Niklas et al
(1995).
The ionizing flux is produced almost exclusively by stars with M > 10 M⊙ , so SFRs derived from
this method are especially sensitive to the form of the IMF. Adopting the Scalo (1986) IMF, for
example, yields SFRs that are ∼3 times higher than derived with a Salpeter IMF. Fortunately,
8
the Hα equivalent widths and broadband colors of galaxies are very sensitive to the slope of the
IMF over the mass range 1–30 M⊙ , and these can be used to constrain the IMF slope (Kennicutt
1983a, Kennicutt et al 1994). The properties of normal disks are well fitted by a Salpeter IMF
(or by a Scalo function with Salpeter slope above 1 M⊙ ), consistent with observations of resolved
stellar populations in nearby galaxies (e.g. Massey 1998). As with the UV continuum method, it
is important when applying published SFRs to take proper account of the IMF that was assumed.
2.4
Forbidden Lines
The Hα emission line is redshifted out of the visible window beyond z∼0.5, so there is considerable
interest in calibrating bluer emission lines as quantitative SFR tracers. Unfortunately the integrated
strengths of Hβ and the higher order Balmer emission lines are poor SFR diagnostics, because the
lines are weak and stellar absorption more strongly influences the emission-line fluxes. These lines in
fact are rarely seen in emission at all in the integrated spectra of galaxies earlier than Sc (Kennicutt
1992a, also see Figure 1).
The strongest emission feature in the blue is the [OII]λ3727 forbidden-line doublet. The luminosities
of forbidden lines are not directly coupled to the ionizing luminosity, and their excitation is sensitive
to abundance and the ionization state of the gas. However the excitation of [OII] is sufficiently well
behaved that it can be calibrated empirically (through Hα) as a quantitative SFR tracer. Even
this indirect calibration is extremely useful for lookback studies of distant galaxies, because [OII]
can be observed in the visible out to redshifts z ∼ 1.6, and it has been measured in several large
samples of faint galaxies (Cowie et al 1996, 1997, Ellis 1997, and references therein).
Calibrations of SFRs in terms of [OII] luminosity have been published by Gallagher et al (1989),
based on large-aperture spectrophotometry of 75 blue irregular galaxies, and by Kennicutt (1992a),
using integrated spectrophotometry of 90 normal and peculiar galaxies. When converted to the
same IMF and Hα calibration the resulting SFR scales differ by a factor of 1.57, reflecting excitation
differences in the two samples. Adopting the average of these calibrations yields:
SF R (M⊙ yr −1 ) = (1.4 ± 0.4) × 10−41 L[OII] (ergs s−1 ),
(3)
where the uncertainty indicates the range between blue emission-line galaxies (lower limit) and
samples of more luminous spiral and irregular galaxies (upper limit). As with Equations 1 and
2, the observed luminosities must be corrected for extinction, in this case the extinction at Hα,
because of the manner in which the [OII] fluxes were calibrated.
The SFRs derived from [OII] are less precise than from Hα, because the mean [OII]/Hα ratios in
individual galaxies vary considerably, over 0.5–1.0 dex in the Gallagher et al (1989) and Kennicutt
(1992a) samples, respectively. The [OII]-derived SFRs may also be prone to systematic errors from
extinction and variations in the diffuse gas fraction. The excitation of [OII] is especially high in
the diffuse ionized gas in starburst galaxies (Hunter & Gallagher 1990, Hunter 1994, Martin 1997),
enough to more than double the L[OII]/SFR ratio in the integrated spectrum (Kennicutt 1992a).
On the other hand, metal abundance has a relatively small effect on the [OII] calibration, over most
9
of the abundance range of interest (0.05 Z⊙ ≤ Z ≤ 1 Z⊙ ). Overall the [OII] lines provide a very
useful estimate of the systematics of SFRs in samples of distant galaxies, and are especially useful
as a consistency check on SFRs derived in other ways.
2.5
Far-Infrared Continuum
A significant fraction of the bolometric luminosity of a galaxy is absorbed by interstellar dust and
re-emitted in the thermal IR, at wavelengths of roughly 10–300 µm. The absorption cross section
of the dust is strongly peaked in the ultraviolet, so in principle the FIR emission can be a sensitive
tracer of the young stellar population and SFR. The IRAS survey provides FIR fluxes for over
30,000 galaxies (Moshir et al 1992), offering a rich reward to those who can calibrate an accurate
SFR scale from the 10–100 µm FIR emission.
The efficacy of the FIR luminosity as a SFR tracer depends on the contribution of young stars to
heating of the dust, and on the optical depth of the dust in the star forming regions. The simplest
physical situation is one in which young stars dominate the radiation field thoughout the UV–visible,
and the dust opacity is high everywhere, in which case the FIR luminosity measures the bolometric
luminosity of the starburst. In such a limiting case the FIR luminosity is the ultimate SFR tracer,
providing what is essentially a calorimetric measure of the SFR. Such conditions roughly hold in
the dense circumnuclear starbursts that power many IR-luminous galaxies.
The physical situation is more complex in the disks of normal galaxies, however (e.g. Lonsdale &
Helou 1987, Rowan-Robinson & Crawford 1989, Cox & Mezger 1989). The FIR spectra of galaxies
contain both a “warm” component associated with dust around young star forming regions (λ̄ ∼
60µm), and a cooler “infrared cirrus” component (λ̄ ≥ 100µm) which is associated with more
extended dust heated by the interstellar radiation field. In blue galaxies, both spectral components
may be dominated by young stars, but in red galaxies, where the composite stellar continuum drops
off steeply in the blue, dust heating from the visible spectra of older stars may be very important.
The relation of the global FIR emission of galaxies to the SFR has been a controversial subject.
In late-type star forming galaxies, where dust heating from young stars is expected to dominate
the 40–120µm emission, the FIR luminosity correlates with other SFR tracers such as the UV
continuum and Hα luminosities (e.g. Lonsdale & Helou 1987, Sauvage & Thuan 1992, Buat & Xu
1996). However, early-type (S0–Sab) galaxies often exhibit high FIR luminosities but much cooler,
cirrus-dominated emission. This emission has usually been attributed to dust heating from the
general stellar radiation field, including the visible radiation from older stars (Lonsdale & Helou
1987, Buat & Deharveng 1988, Rowan-Robinson & Crawford 1989, Sauvage & Thuan 1992, 1994,
Walterbos & Greenawalt 1996). This interpretation is supported by anomalously low UV and Hα
emission (relative to the FIR luminosity) in these galaxies. However Devereux & Young (1990) and
Devereux & Hameed (1997) have argued that young stars dominate the 40–120µm emission in all
of these galaxies, so that the FIR emission directly traces the SFR. They have provided convincing
evidence that young stars are an important source of FIR luminosity in at least some early-type
galaxies, including barred galaxies with strong nuclear starbursts and some unusually blue objects
(Section 4). On the other hand, many early-type galaxies show no independent evidence of high
10
SFRs, suggesting that the older stars or active galactic nuclei (AGNs) are responsible for much of
the FIR emission. The Space Infrared Telescope Facility, scheduled for launch early in the next
decade, should provide high-resolution FIR images of nearby galaxies and clarify the relationship
between the SFR and IR emission in these galaxies.
The ambiguities discussed above affect the calibration of SFRs in terms of FIR luminosity, and
there probably is no single calibration that applies to all galaxy types. However the FIR emission
should provide an excellent measure of the SFR in dusty circumnuclear starbursts. The SFR vs
LF IR conversion is derived using synthesis models as described earlier. In the optically thick limit,
it is only necessary to model the bolometric luminosity of the stellar population. The greatest
uncertainty in this case is the adoption of an appropriate age for the stellar population; this may
be dictated by the timescale of the starburst itself or by the timescale for the dispersal of the dust
(so the τ ≫1 approximation no longer holds). Calibrations have been published by several authors
under different assumptions about the star formation timescale (e.g. Hunter et al 1986, Lehnert &
Heckman 1996, Meurer et al 1997, Kennicutt 1998). Applying the models of Leitherer & Heckman
(1995) for continuous bursts of age 10–100 Myr, and adopting the IMF in this paper yields the
relation (Kennicutt 1998):
SFR (M⊙ yr −1 ) = 4.5 × 10−44 LF IR (ergs s−1 ) (starbursts),
(4)
where LF IR refers to the infrared luminosity integrated over the full mid and far-IR spectrum
(8–1000 µm), though for starbursts most of this emission will fall in the 10–120µm region (readers
should beware that the definition of LF IR varies in the literature). Most of the other published
calibrations lie within ±30% of Equation 4. Strictly speaking, the relation given above applies only
to starbursts with ages less than 108 years, where the approximations applied are valid. In more
quiescent normal star forming galaxies, the relation will be more complicated; the contribution of
dust heating from old stars will tend to lower the effective coefficient in equation [4], whereas the
lower optical depth of the dust will tend to increase the coefficient. In such cases, it is probably
better to rely on an empirical calibration of SFR/LF IR , based on other methods. For example,
−44 , valid for galaxies of type Sb and later only,
Buat & Xu (1996) derive a coefficient of 8+8
−3 × 10
based on IRAS and UV flux measurements of 152 disk galaxies. The FIR luminosities share the
same IMF sensitivity as the other direct star formation tracers, and it is important to be consistent
when comparing results from different sources.
3
DISK STAR FORMATION
The techniques described above have been used to measure SFRs in hundreds of nearby galaxies,
and these have enabled us to delineate the main trends in SFRs and star formation histories along
the Hubble sequence. Although it is customary to analyze the integrated SFRs of galaxies, taken
as a whole, large-scale star formation takes place in two very distinct physical environments: one
in the extended disks of spiral and irregular galaxies, the other in compact, dense gas disks in the
centers of galaxies. Both regimes are significant contributors to the total star formation in the
local universe, but they are traced at different wavelengths and follow completely different patterns
11
along the Hubble sequence. Consequently I will discuss the disk and circumnuclear star formation
properties of galaxies separately.
3.1
Global SFRs Along the Hubble Sequence
Comprehensive analyses of the global SFRs of galaxies have been carried out using Hα surveys
(Kennicutt 1983a, Gallagher et al 1984, Caldwell et al 1991, 1994, Kennicutt et al 1994, Young et
al 1996), UV continuum surveys (Donas et al 1987, Deharveng et al 1994), FIR data (Sauvage &
Thuan 1992, Walterbos & Greenawalt 1996, Tomita et al 1996, Devereux & Hameed 1997), and
multi-wavelength surveys (Gavazzi & Scodeggio 1996, Gavazzi et al 1996). The absolute SFRs in
galaxies, expressed in terms of the total mass of stars formed per year, show an enormous range,
from virtually zero in gas-poor elliptical, S0, and dwarf galaxies to ∼20 M⊙ yr−1 in gas-rich spirals.
Much larger global SFRs, up to ∼100 M⊙ yr−1 , can be found in optically-selected starburst galaxies,
and SFRs as high as 1000 M⊙ yr−1 may be reached in the most luminous IR starburst galaxies
(Section 4). The highest SFRs are associated almost uniquely with strong tidal interactions and
mergers.
Part of the large dynamic range in absolute SFRs simply reflects the enormous range in galaxy
masses, so it is more illuminating to examine the range in relative SFRs, normalized per unit mass
or luminosity. This is illustrated in Figure 3, which shows the distribution of Hα+[NII] equivalent
widths (EWs) in a sample of 227 nearby bright galaxies (BT < 13), subdivided by Hubble type.
The data were taken from the photometric surveys of Kennicutt & Kent (1983) and Romanishin
(1990). The measurements include the Hα and the neighboring [NII] lines; corrections for [NII]
contamination are applied when determining the SFRs. The EW is defined as the emission-line
luminosity normalized to the adjacent continuum flux, and hence it is a measure of the SFR per
unit (red) luminosity.
Figure 3 shows a range of more than two orders of magnitude in the SFR per unit luminosity. The
EWs show a strong dependence on Hubble type, increasing from zero in E/S0 galaxies (within the
observational errors) to 20–150 Å in late-type spiral and irregular galaxies. When expressed in terms
of absolute SFRs, this corresponds to range of 0–10 M⊙ yr−1 for an L∗ galaxy (roughly comparable
in luminosity to the Milky Way). The SFR measured in this way increases by approximately a
factor of 20 between types Sa and Sc (Caldwell et al 1991, Kennicutt et al 1994). SFRs derived
from the UV continuum and broadband visible colors show comparable behavior (e.g. Larson &
Tinsley 1978, Donas et al 1987, Buat et al 1989, Deharveng et al 1994).
High-resolution imaging of individual galaxies reveals that the changes in the disk SFR along the
Hubble sequence are produced in roughly equal parts by an increase in the total number of star
forming regions per unit mass or area, and an increase in the characteristic masses of individual
regions (Kennicutt et al 1989a, Caldwell et al 1991, Bresolin & Kennicutt 1997). These trends
are seen both in the Hα luminosities of the HII regions as well as in the continuum luminosity
functions of the embedded OB associations (Bresolin & Kennicutt 1997). A typical OB star in an
Sa galaxy forms in a cluster containing only a few massive stars, whereas an average massive star
in a large Sc or Irr galaxy forms in a giant HII/OB association containing hundreds or thousands
12
Figure 3: Distribution of integrated Hα+[NII] emission-line equivalent widths for a large sample of
nearby spiral galaxies, subdivided by Hubble type and bar morphology. The right axis scale shows
corresponding values of the stellar birthrate parameter b, which is the ratio of the present SFR to
that averaged over the past, as described in Section 5.1.
of OB stars. These differences in clustering properties of the massive stars may strongly influence
the structure and dynamics of the interstellar medium (ISM) along the Hubble sequence (Norman
& Ikeuchi 1989, Heiles 1990).
Although there is a strong trend in the average SFRs with Hubble type, a dispersion of a factor of
ten is present in SFRs among galaxies of the same type. The scatter is much larger than would be
expected from observational errors or extinction effects, so most of it must reflect real variations
in the SFR. Several factors contribute to the SFR variations, including variations in gas content,
nuclear emission, interactions, and possibly short-term variations in the SFR within individual
objects. Although the absolute SFR varies considerably among spirals (types Sa and later), some
level of massive star formation is always observed in deep Hα images (Caldwell et al 1991). However
many of the earliest disk galaxies (S0–S0/a) show no detectable star formation at all. Caldwell et
al (1994) obtained deep Fabry-Perot Hα imaging of 8 S0–S0/a galaxies, and detected HII regions
in only 3 objects. The total SFRs in the latter galaxies are very low, <0.01 M⊙ yr−1 , and the
13
upper limits on the other 4 galaxies rule out HII regions fainter than those of the Orion nebula.
On the other hand, Hα surveys of HI-rich S0 galaxies by Pogge & Eskridge (1987, 1993) reveal a
higher fraction of disk and/or circumnuclear star forming regions, emphasizing the heterogeneous
star formation properties of these galaxies. Thronson et al (1989) reached similar conclusions based
on an analysis of IRAS observations of S0 galaxies.
The relative SFRs can also be parametrized in terms of the mean SFR per unit disk area. This
has the advantage of avoiding any effect of bulge contamination on total luminosities (which biases
the EW distributions). Analyses of the SFR surface density distributions have been published
by Deharveng et al (1994), based on UV continuum observations, and by Ryder (1993), Ryder
& Dopita (1994), and Young et al (1996), based on Hα observations. The average SFR surface
densities show a similar increase with Hubble type, but the magnitude of the change is noticeably
weaker than is seen in SFRs per unit luminosity (e.g. Figure 3), and the dispersion among galaxies
of the same type is larger (see below). The stronger type dependence in the Hα EWs (see Figure
3) is partly due to the effects of bulge contamination, which exaggerate the change in disk EWs by
a factor of two between types Sa–Sc (Kennicutt et al 1994), but the change in disk EWs with type
is still nearly twice as large as the comparable trend in SFR per unit area (Young et al 1996). The
difference reflects the tendency for the late-type spirals to have somewhat more extended (i.e. lower
surface brightness) star forming disks than the early-type spirals, at least in these samples. This
comparison demonstrates the danger in applying the term SFR too loosely when characterizing the
systematic behavior of star formation across the Hubble sequence, because the quantitative trends
are dependent on the manner in which the SFR is defined. Generally speaking, a parameter that
scales with the SFR per unit mass (e.g. the Hα equivalent width) is most relevant to interpreting the
evolutionary properties of disks, whereas the SFR per unit area is more relevant to parametrizing
the dependence of the SFR on gas density in disks.
Similar comparisons can be made for the FIR properties of disk galaxies, and these show considerably weaker trends with Hubble type (Devereux & Young 1991, Tomita et al 1996, Devereux &
Hameed 1997). This is illustrated in Figure 4, which shows the distributions of LF IR /LH from a
sample of nearby galaxies studied by Devereux & Hameed (1997). Since the near-IR H-band luminosity is a good indicator of the total stellar mass, the LF IR /LH ratio provides an approximate
measure of the FIR emission normalized to the mass of the parent galaxy. Figure 4 shows the
expected trend toward stronger FIR emission with later Hubble type, but the trend is considerably
weaker, in the sense that early-type galaxies show much higher FIR luminosities than would be
expected given their UV-visible spectra. Comparisons of LF IR /LB distributions show almost no
dependence on Hubble type at all (Isobe & Feigelson 1992, Tomita et al 1996, Devereux & Hameed
1997), but this is misleading because the B-band luminosity itself correlates with the SFR (see
Figure 2).
The inconsistencies between the FIR and UV–visible properties of spiral galaxies appear to be
due to a combination of effects (as mentioned above in Section 2.5). In at least some early-type
spirals, the strong FIR emission is produced by luminous, dusty star forming regions, usually
concentrated in the central regions of barred spiral galaxies (Devereux 1987, Devereux & Hameed
1997). This exposes an important bias in the visible and UV-based studies of SFRs in galaxies, in
that they often do not take into account the substantial star formation in the dusty nuclear regions,
14
Figure 4: Distributions of 40- to 120-µm infrared luminosity for nearby galaxies, normalized to nearinfrared H luminosity, as a function of Hubble type. Adapted from Devereux & Hameed (1997),
with elliptical and irregular galaxies excluded.
which can dominate the global SFR in an early-type galaxy. Devereux & Hameed emphasize the
importance of observing a sufficiently large and diverse sample of early-type galaxies, in order to
fully characterize the range of star formation properties. However it is also likely that much of
the excess FIR emission in early-type spirals is unrelated to star formation, reflecting instead the
effects of dust heating from evolved stellar populations (Section 2.5). Radiative transfer modelling
by Walterbos & Greenawalt (1996) demonstrates that this effect can readily account for the trends
seen in Figure 4.
The interpretation in the remainder of this review is based on the SFR trends revealed by the Hα,
UV continuum, broadband colors, and integrated spectra, which are consistent with a common
evolutionary picture of the Hubble sequence. However it is important to bear in mind that this
picture applies only to the extended, extranuclear star formation in spiral and irregular disks. The
circumnuclear star formation follows quite different patterns, as discussed in Section 4.2.
15
3.2
Dependence of SFRs on Gas Content
The strong trends in disk SFRs that characterize the Hubble sequence presumably arise from more
fundamental relationships between the global SFR and other physical properties of galaxies, such
as their gas contents or dynamical structure. The physical regulation of the SFR is a mature
subject in its own right, and a full discussion is beyond the scope of this review. However it is very
instructive to examine the global relationships between the disk-averaged SFRs and gas densities
of galaxies, because they reveal important insights into the physical nature of the star formation
sequence, and they serve to quantify the range of physical conditions and evolutionary properties
of disks.
Comparisons of the large-scale SFRs and gas contents of galaxies have been carried out by several
authors, most recently Buat et al (1989), Kennicutt (1989), Buat (1992), Boselli (1994), Deharveng
et al (1994), Boselli et al (1995) and Kennicutt (1998). Figure 5 shows the relationship between
the disk-averaged SFR surface density ΣSF R and average total (atomic plus molecular) gas density
Σgas , for a sample of 61 normal spiral galaxies with Hα, HI, and CO observations (Kennicutt 1998).
The SFRs were derived from extinction-corrected Hα fluxes, using the SFR calibration in Equation
2. The surface densities were averaged within the corrected optical radius R0 , as taken from the
Second Reference Catalog of Bright Galaxies (de Vaucouleurs et al 1976).
Figure 5 shows that disks possess large ranges in both the mean gas density (factor of 20–30) and
mean SFR surface density (factor of 100). The data points are coded by galaxy type, and they
show that both the gas and SFR densities are correlated with Hubble type on average, but with
large variations among galaxies of a given type. In addition, there is an underlying correlation
between SFR and gas density that is largely independent of galaxy type. This shows that much of
the scatter in SFRs among galaxies of the same type can be attributed to an underlying dispersion
in gas contents. The data can be fitted to a Schmidt (1959) law of the form ΣSF R = A ΣN
gas . The
best fitting slope N ranges from 1.4 for a conventional least squares fit (minimizing errors in SFRs
only) to N =2.4 for a bivariate regression, as shown by the solid lines in Figure 5. Values of N in
the range 0.9–1.7 have been derived by previous workers, based on SFRs derived from Hα, UV, and
FIR data (Buat et al 1989, Kennicutt 1989, Buat 1992, Deharveng et al 1994). The scatter in SFRs
at a given gas density is large, and most of this dispersion is probably introduced by averaging the
SFRs and gas densities over a large dynamic range of local densities within the individual disks
(Kennicutt 1989, 1998).
Figure 5 also contains information on the typical global efficiencies of star formation and gas
consumption time scales in disks. The dashed and dotted lines indicate constant, disk-averaged
efficiencies of 1%, 10%, and 100% per 108 years. The average value for these galaxies is 4.8%,
meaning that the average disk converts 4.8% of its gas (within the radius of the optical disk) every
108 years. Since the typical gas mass fraction in these disk is about 20%, this implies that stellar
mass of the disk grows by about 1% per 108 years, i.e. the time scale for building the disk (at the
present rate) is comparable to the Hubble time. The efficiencies can also be expressed in terms
of the average gas depletion timescale, which for this sample is 2.1 Gyr. Recycling of interstellar
gas from stars extends the actual time scale for gas depletion by factors of 2–3 (Ostriker & Thuan
1975, Kennicutt et al 1994).
16
Figure 5: Correlation between disk-averaged SFR per unit area and average gas surface density,
for 61 normal disk galaxies. Symbols are coded by Hubble type: Sa–Sab (open triangles); Sb–Sbc
(open circles); Sc–Sd (solid points); Irr (cross). The dashed and dotted lines show lines of constant
global star formation efficiency. The error bars indicate the typical uncertainties for a given galaxy,
including systematic errors.
3.3
Other Global Influences on the SFR
What other global properties of a galaxy influence its SFR? One might plausibly expect the mass,
bar structure, spiral arm structure, or environment to be important, and empirical information on
all of these are available.
3.3.1 LUMINOSITY AND MASS Gavazzi & Scodeggio (1996) and Gavazzi et al (1996) have
compiled UV, visible, and near-IR photometry for over 900 nearby galaxies, and they found an anticorrelation between the SFR per unit mass and the galaxy luminosity, as indicated by broadband
colors and Hα EWs. At least part of this trend seems to reflect the same dependence of SFR
on Hubble type discussed above, but a mass dependence is also observed among galaxies of the
same Hubble type. It is interesting that there is considerable overlap between the color-luminosity
relations of different spiral types, which suggests that part of the trends that are attributed to
17
morphological type may be more fundamentally related to total mass. A strong correlation between
B–H color and galaxy luminosity or linewidth has been discussed previously by Tully et al (1982)
and Wyse (1983). The trends seem to be especially strong for redder colors, which are more closely
tied to the star formation history and mean metallicity than the current SFR. More data are needed
to fully disentangle the effects of galaxy type and mass, both for the SFR and the star formation
history.
3.3.2 BARS Stellar bars can strongly perturb the gas flows in disks, and trigger nuclear star
formation (see next section), but they do not appear to significantly affect the total disk SFRs.
Figure 3 plots the Hα EW distributions separately for normal (SA and SAB) and barred (SB)
spirals, as classified in the Second Reference Catalog of Bright Galaxies. There is no significant
difference in the EW distributions (except possibly for the Sa/SBa galaxies), which suggests that
the global effect of a bar on the disk SFR is unimportant. Ryder & Dopita (1994) reached the same
conclusion based on Hα observations of 24 southern galaxies.
Tomita et al (1996) have carried out a similar comparison of FIR emission, based on IRAS data
and broadband photometry for 139 normal spirals and 260 barred Sa–Sc galaxies. They compared
the distributions of LF IR/LB ratios for Sa/SBa, Sb/SBb, and Sc/SBc galaxies, and concluded
that there is no significant correlation with bar structure, consistent with the Hα results. There
is evidence for a slight excess in FIR emission in SBa galaxies, which could reflect bar-triggered
circumnuclear star formation in some of the galaxies, though the statistical significance of the result
is marginal (Tomita et al 1996).
Recent work by Martinet & Friedli (1997) suggests that influence of bars on the global SFR may
not be as simple as suggested above. They analyzed Hα and FIR-based SFRs for a sample of 32
late-type barred galaxies, and found a correlation between SFR and the strength and length of the
bar. This suggests that large samples are needed to study the effects of bars on the large-scale SFR,
and that the structural properties of the bars themselves need to be incorporated in the analysis.
3.3.3 SPIRAL ARM STRUCTURE Similar tests have been carried out to explore whether
a strong grand-design spiral structure enhances the global SFR. Elmegreen & Elmegreen (1986)
compared UV and visible broadband colors and Hα EWs for galaxies they classified as granddesign (strong two-armed spiral patterns) and flocculent (ill-defined, patchy spiral arms), and they
found no significant difference in SFRs. McCall & Schmidt (1986) compared the supernova rates
in grand-design and flocculent spirals, and drew similar conclusions. Grand-design spiral galaxies
show strong local enhancements of star formation in their spiral arms (e.g. Cepa & Beckman 1990,
Knapen et al 1992), so the absence of a corresponding excess in their total SFRs suggests that the
primary effect of the spiral density wave is to concentrate star formation in the arms, but not to
increase the global efficiency of star formation.
3.3.4 GALAXY-GALAXY INTERACTIONS Given the modest effects of internal disk structure
on global SFRs, it is perhaps somewhat surprising that external environmental influences can have
much stronger effects on the SFR. The most important influences by far are tidal interactions.
Numerous studies of the global Hα and FIR emission of interacting and merging galaxies have
shown strong excess star formation (e.g. Bushouse 1987, Kennicutt et al 1987, Bushouse et al
1988, Telesco et al 1988, Xu & Sulentic 1991, Liu & Kennicutt 1995). The degree of the SFR
18
enhancement is highly variable, ranging from zero in gas-poor galaxies to on the order of 10–100
times in extreme case. The average enhancement in SFR over large samples is a factor of 2–3.
Much larger enhancements are often seen in the circumnuclear regions of strongly interacting and
merging systems (see next section). This subject is reviewed in depth in Kennicutt et al (1998).
3.3.5 CLUSTER ENVIRONMENT There is evidence that cluster environment systematically
alters the star formation properties of galaxies, independently of the well-known density-morphology
relation (Dressler 1984). Many spiral galaxies located in rich clusters exhibit significant atomic gas
deficiencies (Haynes et al 1984, Warmels 1988, Cayatte et al 1994), which presumably are the
result of ram pressure stripping from the intercluster medium, combined with tidal stripping from
interactions with other galaxies and the cluster potential. In extreme cases one would expect the
gas removal to affect the SFRs as well. Kennicutt (1983b) compared Hα EWs of 26 late-type spirals
in the Virgo cluster core with the field sample of Kennicutt & Kent (1983) and found evidence for
a 50% lower SFR in Virgo, comparable to the level of HI deficiency. The UV observations of the
cluster Abell 1367 by Donas et al (1990) also show evidence for lower SFRs. However subsequent
studies of the Coma, Cancer, and A1367 clusters by Kennicutt et al (1984) and Gavazzi et al (1991)
showed no reduction in the average SFRs, and if anything a higher number of strong emission-line
galaxies.
A comprehensive Hα survey of 8 Abell clusters by Moss & Whittle (1993) suggests that the effects
of cluster environoment on global star formation activity are quite complex. They found a 37–46%
lower Hα detection rate among Sb, Sc, and irregular galaxies in the clusters, but a 50% higher
detection rate among Sa–Sab galaxies. They argue that these results arise from a combination of
competing effects, including reduced star formation from gas stripping as well as enhanced star
formation triggered by tidal interactions. Ram-pressure induced star formation may also be taking
place in a few objects (Gavazzi & Jaffe 1985).
4
CIRCUMNUCLEAR STAR FORMATION
AND STARBURSTS
It has been known from the early photographic work of Morgan (1958) and Sérsic & Pastoriza
(1967) that the circumnuclear regions of many spiral galaxies harbor luminous star forming regions,
with properties that are largely decoupled from those of the more extended star forming disks.
Subsequent spectroscopic surveys revealed numerous examples of bright emission-line nuclei with
spectra resembling those of HII regions (e.g. Heckman et al 1980, Stauffer 1982, Balzano 1983,
Keel 1983). The most luminous of these were dubbed “starbursts” by Weedman et al (1981).
The opening of the mid-IR and FIR regions fully revealed the distinctive nature of the nuclear
star formation (e.g. Rieke & Low 1972, Harper & Low 1973, Rieke & Lebofsky 1978, Telesco &
Harper 1980). The IRAS survey led to the discovery of large numbers of ultraluminous star forming
galaxies (Soifer et al 1987). This subject has grown into a major subfield of its own, which has been
thoroughly reviewed elsewhere in this series (Soifer et al 1987, Telesco 1988, Sanders & Mirabel
1996). The discussion here focusses on the range of star formation properties of the nuclear regions,
and the patterns in these properties along the Hubble sequence.
19
4.1
SFRs and Physical Properties
Comprehensive surveys of the star formation properties of galactic nuclei have been carried out
using emission-line spectroscopy in the visible (Stauffer 1982, Keel 1983, Kennicutt et al 1989b,
Ho et al 1997a, b) and mid-IR photometry (Rieke & Lebofsky 1978, Scoville et al 1983, Devereux
et al 1987, Devereux 1987, Giuricin et al 1994). Nuclear emission spectra with HII region-like line
ratios are found in 42% of bright spirals (BT < 12.5), with the fraction increasing from 8% in
S0 galaxies (and virtually zero in elliptical galaxies) to 80% in Sc–Im galaxies (Ho et al 1997a).
These fractions are lower limits, especially in early-type spirals, because the star formation often
is masked by a LINER or Seyfert nucleus. Similar detection fractions are found in 10µm surveys
of optically-selected spiral galaxies, but with a stronger weighting toward early Hubble types. The
nuclear SFRs implied by the Hα and IR fluxes span a large range, from a lower detection limit of
∼10−4 M⊙ yr−1 to well over 100 M⊙ yr−1 in the most luminous IR galaxies.
The physical character of the nuclear star forming regions changes dramatically over this spectrum
of SFRs. The nuclear SFRs in most galaxies are quite modest, averaging ∼0.1 M⊙ yr−1 (median
0.02 M⊙ yr−1 ) in the Hα sample of Ho et al (1997a), and ∼0.2 M⊙ yr−1 in the (optically selected)
10µm samples of Scoville et al (1983) and Devereux et al (1987). Given the different selection criteria
and completeness levels in these surveys, the SFRs are reasonably consistent with each other, and
this suggests that the nuclear star formation at the low end of the SFR spectrum typically occurs
in moderately obscured regions (AHα ∼0–3 mag) that are not physically dissimilar from normal
disk HII regions (Kennicutt et al 1989b, Ho et al 1997a).
However the IR observations also reveal a population of more luminous regions, with LF IR ∼ 1010 –
1013 L⊙ , and corresponding SFRs of order 1–1000 M⊙ yr−1 (Rieke & Low 1972, Scoville et al
1983, Joseph & Wright 1985, Devereux 1987). Such high SFRs are not seen in optically-selected
samples, mainly because the luminous starbursts are uniquely associated with dense molecular gas
disks (Young & Scoville 1991 and references therein), and for normal gas-to-dust ratios, one expects
visible extinctions of several magnitudes or higher. The remainder of this section will focus on these
luminous nuclear starbursts, because they represent a star formation regime that is distinct from
the more extended star formation in disks, and because these bursts often dominate the total SFRs
in their parent galaxies.
The IRAS all-sky survey provided the first comprehensive picture of this upper extreme in the SFR
spectrum. Figure 6 shows a comparison of the total 8–1000 µm luminosities (as derived from IRAS)
and total molecular gas masses for 87 IR-luminous galaxies, taken from the surveys of Tinney et al
(1990) and Sanders et al (1991). Tinney et al’s sample (open circles) includes many luminous but
otherwise normal star forming galaxies, while Sanders et al’s brighter sample (solid points) mainly
comprises starburst galaxies and a few AGNs. Strictly speaking these measurements cannot be
applied to infer the nuclear SFRs of the galaxies, because they are low-resolution measurements
and the samples are heterogeneous. However circumnuclear star formation sufficiently dominates
the properties of the luminous infrared galaxies (e.g. Veilleux et al 1995, Lutz et al 1996) that Figure
6 (solid points) provides a representative indication of the range of SFRs in these IRAS-selected
samples.
20
Figure 6: Relationship between integrated far-infrared (FIR) luminosity and molecular gas mass
for bright IR galaxies, from Tinney et al (1990; open circles) and a more luminous sample from
Sanders et al (1991; solid points). The solid line shows the typical L/M ratio for galaxies similar
to the Milky Way. The dashed line shows the approximate limiting luminosity for a galaxy forming
stars with 100% efficiency on a dynamical timescale, as described in the text.
The most distinctive feature in Figure 6 is the range of infrared luminosities. The lower range
overlaps with the luminosity function of normal galaxies (the lower limit of 1010 L⊙ is the sample
definition cutoff), but the population of infrared galaxies extends upward to >1012.5 L⊙ . This
would imply SFRs of up to 500 M⊙ yr−1 (Equation 4), if starbursts are primarily responsible for
the dust heating, about 20 times larger than the highest SFRs observed in normal galaxies. Figure
6 also shows that the luminous IR galaxies are associated with unusally high molecular gas masses,
which partly accounts for the high SFRs. However the typical SFR per unit gas mass is much
higher than in normal disks; the solid line in Figure 6 shows the typical L/M ratio for normal
galaxies, and the efficiencies in the IR galaxies are higher by factors of 2–30 (Young et al 1986,
Solomon & Sage 1988, Sanders et al 1991). The H2 masses used here have been derived using a
standard Galactic H2 /CO conversion ratio, and if the actual conversion factor in the IRAS galaxies
is lower, as is suggested by several lines of evidence, the contrast in star formation efficiencies would
be even larger (e.g. Downes et al 1993, Aalto et al 1994, Solomon et al 1997).
High-resolution IR photometry and imaging of the luminous infrared galaxies reveals that the bulk
of the luminosity originates in compact circumnuclear regions (e.g. Wright et al 1988, Carico et
al 1990, Telesco et al 1993, Smith & Harvey 1996, and references therein). Likewise, CO interferometric observations show that a large fraction of the molecular gas is concentrated in central
disks, with typical radii on the order of 0.1–1 kpc, and implied surface densities on the order of
102 –105 M⊙ pc−2 (Young & Scoville 1991, Scoville et al 1994, Sanders & Mirabel 1996). Less mas21
Table 1: Star Formation in Disks and Nuclei of Galaxies
Property
Radius
SFR
Bolometric Luminosity
Gas Mass
Star Formation Timescale
Gas Density
Optical Depth (0.5 µm)
SFR Density
Dominant Mode
Type Dependence?
Bar Dependence?
Spiral Structure Dependence?
Interactions Dependence?
Cluster Dependence?
Redshift Dependence?
Spiral Disks
Circumnuclear Regions
1 − 30 kpc
0 − 20 M⊙ yr−1
106 − 1011 L⊙
108 − 1011 M⊙
1 − 50 Gyr
1 − 100 M⊙ pc−2
0−2
0 − 0.1 M⊙ yr−1 kpc−2
steady state
0.2 − 2 kpc
0 − 1000 M⊙ yr−1
106 − 1013 L⊙
106 − 1011 M⊙
0.1 − 1 Gyr
102 − 105 M⊙ pc−2
1 − 1000
1 − 1000 M⊙ yr−1 kpc−2
steady state + burst
strong
weak/none
weak/none
moderate
moderate/weak
strong
weak/none
strong
weak/none
strong
?
?
sive disks but with similar gas and SFR surface densities are associated with the infrared-bright
nuclei of spiral galaxies (e.g. Young & Scoville 1991, Telesco et al 1993, Scoville et al 1994, Smith &
Harvey 1996, Rubin et al 1997). The full spectrum of nuclear starburst regions will be considered
in the remainder of this section.
The physical conditions in the circumnuclear star forming disks are distinct in many respects
from the more extended star forming disks of spiral galaxies, as is summarized in Table 1. The
circumnuclear star formation is especially distinctive in terms of the absolute range in SFRs, the
the much higher spatial concentrations of gas and stars, its burstlike nature (in luminous systems),
and its systematic variation with galaxy type.
The different range of physical conditions in the nuclear starbursts is clearly seen in Figure 7, which
plots the average SFR surface densities and mean molecular surface densities for the circumnuclear
disks of 36 IR-selected starbursts (Kennicutt 1998). The comparison is identical to the SFR–density
plot for spiral disks in Figure 5, except that in this case the SFRs are derived from FIR luminosities
(equation [4]), and only molecular gas densities are plotted. HI observations show that the atomic
gas fractions in these regions are of the order of only a few percent, and can be safely neglected
(Sanders & Mirabel 1996). The SFRs and densities have been averaged over the radius of the
circumnuclear disk, as measured from high-resolution CO or IR maps (see Kennicutt 1998).
22
Figure 7: Correlation between disk-averaged SFR per unit area and average gas surface density,
for 36 infrared-selected circumnuclear starbursts. See Figure 5 for a similar comparison for normal
spiral disks. The dashed and dotted lines show lines of constant star formation conversion efficiency,
with the same notation as in Figure 5. The error bars indicate the typical uncertainties for a given
galaxy, including systematic errors.
Figure 7 shows that the surface densities of gas and star formation in the nuclear starbursts are
1–4 orders of magnitude higher than in spiral disks overall. Densities of this order can be found in
large molecular cloud complexes within spiral disks, of course, but the physical conditions in many
of the nuclear starbursts are extraordinary even by those standards. For example, the typical mean
densities of the largest molecular cloud complexes in M31, M33, and M51 are in the range 40–500
M⊙ pc−2 , which corresponds to the lower range of densities in Figure 7 (Kennicutt 1998). Likewise
the SFR surface densities in the 30 Doradus giant HII region, the most luminous complex in the
Local Group, reaches 100 M⊙ yr−1 kpc−2 only in the central 10 pc core cluster. The corresponding
densities in many of the starbursts exceed these values, over regions as large as a kiloparsec in
radius.
The starbursts follow a relatively well-defined Schmidt law, with index N ∼1.4. The nature of the
star formation law will be discussed further in Section 5, where we examine the SFR vs gas density
23
relation for all of the data taken together. Figure 7 also shows that the characteristic star formation
efficiencies and timescales are quite different in the starbursts. The mean conversion efficiency is
30% per 108 years, 6 times larger than in the spiral disks. Likewise, the gas consumption timescale
is 6 times shorter, about 0.3 Gyr on average. This is hardly surprising— these objects are starbursts
by definition— but Figure 7 serves to quantify the characteristic timescales for the starbursts.
As pointed out by Heckman (1994) and Lehnert & Heckman (1996), the luminous IR galaxies lie
close to the limiting luminosity allowed by stellar energy generation, for a system which converts
all of its gas to stars over a dynamical timescale. For a galaxy with dimensions comparable to
the Milky Way, the minimum timescale for feeding the central starburst is ∼108 years; this is also
consistent with the minimum gas consumption timescale in Figure 7. At the limit of 100% star
formation efficiency over this timescale, the corresponding SFR is trivially:
SFRmax = 100 M⊙ yr−1 (
108 years
Mgas
)
(
).
1010 M⊙
τdyn
(5)
The corresponding maximum bolometric luminosity can be estimated using Equation 4, or by
calculating the maximum nuclear energy release possible from stars over 108 years. The latter is
∼0.01 ǫṁc2 , where ṁ in this case is the SFR, and ǫ is the fraction of the total stellar mass that
is burned in 108 yr. A reasonable value of ǫ for a Salpeter IMF is about 0.05; it could be as high
as 0.2 if the starburst IMF is depleted in low-mass stars (e.g. Rieke et al 1993). Combining these
terms and assuming further that all of the bolometric luminosity is reradiated by the dust yields:
Lmax ∼ 7 × 1011 L⊙ (
Mgas
ǫ
).
)(
10
10 M⊙ 0.05
(6)
Using Equation 4 to convert the SFR to FIR luminosity gives nearly the same coefficient (6 × 1011 ).
This limiting L/M relation is shown by the dashed line in Figure 6, and it lies very close to the
actual upper envelope of the luminous IR galaxies. Given the number of assumptions that went into
Equation 6, this agreement may be partly fortuitous; other physical processes, such as optical depth
effects in the cloud, may also be important in defining the upper luminosity limits (e.g. Downes et
al 1993). However the main intent of this exercise is to illustrate that many of the most extreme
circumnuclear starbursts lie near the physical limit for maximum SFRs in galaxies. Heckman (1994)
extended this argument and derived the maximum SFR for a purely self-gravitating protogalaxy,
and he showed that the most luminous infrared galaxies lie close to this limit as well. Note that
none of these limits apply to AGN-powered galaxies, because the mass consumption requirements
for a given mass are 1–2 orders of magnitude lower.
Taken together, these results reveal the extraordinary character of the most luminous IR starburst
galaxies (Heckman 1994, Scoville et al 1994, Sanders & Mirabel 1996). They represent systems in
which a mass of gas comparable to the entire ISM of a galaxy has been driven into a region of order
1 kpc in size, and this entire ISM is being formed into stars, with almost 100% efficiency, over a
timescale of order 108 years. Such a catastrophic transfer of mass can only take place in a violent
interaction or merger, or perhaps during the initial collapse phase of protogalaxies.
24
4.2
Dependence on Type and Environment
The star formation that takes place in the circumnuclear regions of galaxies also follows quite
different patterns along the Hubble sequence, relative to the more extended star formation in disks.
These distinctions are especially important in early-type galaxies, where the nuclear regions often
dominate the global star formation in their parent galaxies.
4.2.1 HUBBLE TYPE In contrast to the extended star formation in disks, which varies dramatically along the Hubble sequence, circumnuclear star formation is largely decoupled with Hubble
type. Stauffer (1982), Keel (1983), and Ho et al (1997a) have investigated the dependence of nuclear Hα emission in star forming nuclei as a function of galaxy type. The detection frequency of
HII region nuclei is a strong monotonic function of type, increasing from 0% in elliptical galaxies,
to 8% in SO, 22% in Sa, 51% in Sb, and 80% in Sc–Im galaxies (Ho et al 1997a), though these
fractions may be influenced somewhat by AGN contamination. Among the galaxies with nuclear
star formation, the Hα luminosities show the opposite trend; the average extinction-corrected luminosity of HII region nuclei in S0–Sbc galaxies is 9 times higher than in Sc galaxies (Ho et al
1997a). Thus the bulk of the total nuclear star formation in galaxies is weighted toward the earlier
Hubble types, even though the frequency of occurence is higher in the late types.
Similar trends are observed in 10 µm surveys of nearby galaxies (Rieke & Lebofsky 1978, Scoville
et al 1983, Devereux et al 1987, Devereux 1987, Giuricin et al 1994). Interpreting the trends in
nuclear 10µm luminosities by themselves is less straightforward, because the dust can be heated
by active nuclei as well as by star formation, but one can reduce this problem by excluding known
AGNs from the statistics. Devereux et al (1987) analyzed the properties of an optically selected
sample of 191 bright spirals, chosen to lie within or near the distance of the Virgo cluster, and found
that the average nuclear 10µm flux was virtually independent of type and if anything, decreased by
25–30% from types Sa–Sbc to Sc–Scd. An analysis of a larger sample by Giuricin et al (1994) shows
that among galaxies with HII region nuclei (as classified from optical spectra), Sa–Sb nuclei are
1.7 times more luminous at 10µm than Sc galaxies. By contrast the disk SFRs are typically 5–10
times lower in the early-type spirals, so the fractional contribution of the nuclei to the total SFR
increases dramatically in the early-type spirals. The nuclear SFRs in some early-type galaxies are
comparable to the integrated SFRs of late-type spirals (e.g. Devereux 1987, Devereux & Hameed
1997). Thus while luminous nuclear starbursts may occur in across the entire range of spiral host
types (e.g. Rieke & Lebofsky 1978, Devereux 1987), the relative effect is much stronger for the
early-type galaxies; most of the star formation in these galaxies occurs in the circumnuclear regions.
Clearly the physical mechanisms that trigger these nuclear outbursts are largely decoupled from
the global gas contents and SFRs of their parent galaxies.
4.2.2 BAR STRUCTURE These same surveys show that nuclear star formation is strongly
correlated with the presence of strong stellar bars in the parent galaxy. The first clear evidence
came from the photographic work of Sérsic & Pastoriza (1967), who showed that 24% of nearby SB
and SAB galaxies possessed detectable circumnuclear “hotspot” regions, now known to be bright
HII regions and stellar associations. In contrast none of the non-barred galaxies studied showed
evidence for hotspots. This work was followed up by Phillips (1993) who showed that hotspots are
found preferentially in early-type barred galaxies, a tendency noted already by Sérsic & Pastoriza.
25
The effects of bars on the Hα emission from HII region nuclei have been analyzed by Ho et al
(1997b). They found that the incidence of nuclear star formation is higher among the barred
galaxies, but the difference is marginally significant, and no excess is seen among early-type barred
galaxies. However the distributions of Hα luminosities are markedly different, with the barred
galaxies showing an extended tail of bright nuclei that is absent in samples of non-barred galaxies.
This tail extends over a range in Hα luminosities of 3–100 ×1040 ergs s−1 , which corresponds to
SFRs in the range 0.2–8 M⊙ yr−1 . This tail is especially strong in the early-type barred galaxies
(SB0/a–SBbc), where ∼30% of the star forming nuclei have luminosities in this range.
Bars appear to play an especially strong role in triggering the strong IR-luminous starbursts that
are found in early-type spiral galaxies. Hawarden et al (1986) and Dressel (1988) found strong
excess FIR emission in early-type barred spirals, based on IRAS observations, and hypothesized
that this emission was associated with circumnuclear star forming regions. This interpretation was
directly confirmed by Devereux (1987), who detected strong nuclear 10µm emission in 40% of the
early-type barred spirals in his sample. Similar excesses are not seen in samples of late-type barred
galaxies. These results have been confirmed in more extensive later studies by Giuricin et al (1994)
and Huang et al (1996). Although early-type barred galaxies frequently harbor a bright nuclear
starburst, bars are not a necessary condition for such a starburst, as shown by Pompea & Rieke
(1990).
The strong association of nuclear and circumnuclear star formation with bar structure, and the
virtual absence of any other dependence on morphological type, contrasts sharply with the behavior
of the disk SFRs. This implies that the evolution of the circumnuclear region is largely decoupled
from that of the disk at larger radii. The strong distinctions between early-type and late-type
barred galaxies appear to be associated with the structural and dynamical properties of the bars.
Bars in bulge-dominated, early-type spirals tend to be very strong and efficient at transporting
gas from the disk into the central regions, while bars in late-type galaxies are much weaker and
are predicted to be much less efficient in transporting gas (e.g. Athanassoula 1992, Friedli & Benz
1995). All of the results are consistent with a general picture in which the circumnuclear SFRs of
galaxies are determined largely by the rate of gas transport into the nuclear regions.
4.2.3 GALAXY INTERACTIONS AND MERGERS Numerous observations have established a
clear causal link between strong nuclear starbursts and tidal interactions and mergers of galaxies.
Since this subject is reviewed in depth elsewhere (Heckman 1990, 1994, Barnes & Hernquist 1992,
Sanders & Mirabel 1996, Kennicutt et al 1998), only the main results are summarized here.
The evidence for interaction-induced nuclear star formation comes from two types of studies, statistical comparisons of the SFRs in complete samples of interacting and non-interacting galaxies, and
studies of the frequency of interactions and mergers among samples of luminous starburst galaxies.
Keel et al (1985) and Bushouse (1986) showed that the nuclear Hα emission in nearby samples
of interacting spiral galaxies is 3–4 times stronger than that in a control sample of isolated spirals. Groundbased 10–20µm observations of the nuclear regions of interacting and merging galaxies
showed similar or stronger enhancements, depending on how the samples were selected (Lonsdale
et al 1984, Cutri & McAlary 1985, Joseph & Wright 1985, Wright et al 1988). There is an enormous range of SFRs among individual objects. Spatially resolved data also show a stronger central
26
concentration of the star formation in strongly interacting systems (e.g. Bushouse 1987, Kennicutt
et al 1987, Wright et al 1988). Thus while the interactions tend to increase the SFR throughout
galaxies, the effects in the nuclear regions are stronger. This radial concentration is consistent with
the predictions of numerical simulations of interacting and merging systems (Barnes & Hernquist
1992, Mihos & Hernquist 1996, Kennicutt et al 1998).
The complementary approach is to measure the frequencies of interacting systems in samples of
starburst galaxies. The most complete data of this kind come from IRAS, and they show that
the importance of tidal triggering is a strong function of the strength of the starburst, with the
fraction of interactions increasing from 20–30% for LIR <1010 L⊙ to 70–95% for LIR >1012 L⊙ . The
relatively low fraction (Sanders et al 1988, Lawrence et al 1989, Gallimore & Keel 1993, Leech et
al 1994, Sanders & Mirabel 1996). The relatively low fraction for the lower luminosity starbursts
is understandable, because the corresponding SFRs (<1 M⊙ yr−1 ) can be sustained with relatively
modest gas supplies, and can be fed by internal mechanisms such as a strong bar. The most luminous
starbursts, on the other hand, are associated almost exclusively with strong tidal interactions and
mergers. SFRs larger than ∼20 M⊙ yr−1 are rarely observed in isolated galaxies, though some
possible exceptions have been identified by Leech et al (1994). In view of the enormous fueling
requirements for such starbursts (Equations 5 and 6), however, it is perhaps not surprising that an
event as violent as a merger is required. These results underscore the heterogeneous nature of the
starburst galaxy population, and they suggest that several triggering mechanisms are involved in
producing the population.
5
INTERPRETATION AND IMPLICATIONS
FOR GALAXY EVOLUTION
The observations described above can be fitted together into a coherent evolutionary picture of disk
galaxies and the Hubble sequence. This section summarizes the evolutionary implications of these
data, taking into account the distinct patterns seen in the disks and galactic nuclei. It concludes
with a discussion of the critical role of the interstellar gas supply in regulating the SFR, across the
entire range of galaxy types and environments.
5.1
Disk Evolution Along the Hubble Sequence
The strong trends observed in the SFR per unit luminosity along the Hubble sequence mirror
underlying trends in ther past star formation histories of disks (Roberts 1963, Kennicutt 1983a,
Gallagher et al 1984, Sandage 1986, Kennicutt et al 1994). A useful parameter for characterizing
the star formation histories is the ratio of the current SFR to the past SFR averaged over the age
of the disk, denoted b by Scalo (1986). The evolutionary synthesis models discussed in Section 2
provide relations between b and the broadband colors and Hα EWs. Figure 3 shows the distribution
of b (right axis scale) for an Hα-selected sample of galaxies, based on the calibration of Kennicutt
et al (1994). The typical late-type spiral has formed stars at a roughly constant rate (b ∼ 1), which
27
is consistent with direct measurements of the stellar age distribution in the Galactic disk (e.g.,
Scalo 1986). By contrast, early-type spiral galaxies are characterized by rapidly declining SFRs,
with b ∼ 0.01 − 0.1, whereas elliptical and S0 galaxies have essentially ceased forming stars (b = 0).
Although the values of b given above are based solely on synthesis modelling of the Hα equivalent
widths, analysis of the integrated colors and spectra of disks yield similar results (e.g. Kennicutt
1983a, Gallagher et al 1984, Bruzual & Charlot 1993, Kennicutt et al 1994).
The trends in b shown in Figure 3 are based on integrated measurements, so they are affected by
bulge and nuclear contamination, which bias the trends seen along the Hubble sequence. A more
detailed analysis by Kennicutt et al (1994) includes corrections for bulge contamination on the Hα
EWs. The mean value of b (for the disks alone) increases from <0.07 for Sa disks, to 0.3 for Sb
disks and 1.0 for Sc disks. This change is much larger than the change in bulge mass fraction over
the same range of galaxy types, implying that most of the variation in the integrated photometric
properties of spiral galaxies is produced by changes in the star formation histories of the disks,
not in the bulge-to-disk ratio. Variations in bulge-disk structure may be play an important role,
however, in physically driving the evolution of the disks.
As discussed earlier, this picture has been challenged by Devereux & Hameed (1997), based on the
much weaker variation in FIR luminosities along the Hubble sequence. The results of the previous
section provide part of the resolution to this paradox. Many early-type barred spirals harbor
luminous circumnuclear starbursts, with integrated SFRs that can be as high as the disk SFRs in
late-type galaxies. If this nuclear star formation is included, then the interpretation of the Hubble
sequence given above is clearly oversimplistic. For that reason it is important to delineate between
the the nuclear regions and more extended disks when characterizing the evolutionary properties
of galaxies. Much of the remaining difference in interpretations hinges on the nature of the FIR
emission in early-type galaxies, which may not directly trace the SFR in all galaxies.
Although Figure 3 shows a strong change in the average star formation history with galaxy type,
it also shows a large dispersion in b among galaxies of the same type. Some of this must be
due to real long-term variations in star formation history, reflecting the crudeness of the Hubble
classification itself. Similar ranges are seen in the gas contents (Roberts & Haynes 1994), and these
correlate roughly with the SFR and b variations (Figure 5). Short-term variations in the SFR can
also explain part of the dispersion in b. Nuclear starbursts clearly play a role in some galaxies,
especially early-type barred galaxies, and interaction-induced starbursts are observed in a small
percentage of nearby galaxies. Starbursts are thought to be an important, if not the dominant
mode of star formation in low-mass galaxies (e.g. Hunter & Gallagher 1985, Hodge 1989, Ellis
1997), but the role of large-scale starbursts in massive galaxies is less well established, (Bothun
1990, Kennicutt et al 1994, Tomita et al 1996). A definitive answer to this question will probably
come from lookback studies of large samples of disk galaxies.
A schematic illustration of the trends in star formation histories is shown in Figure 8. The left
panel compares the stellar birthrate histories of typical elliptical galaxies (and spiral bulges), and
the disks of Sa, Sb, and Sc galaxies, following Sandage (1986). The curves for the spiral disks are
exponential functions which correspond to the average values of b from Kennicutt et al (1994).
For illustrative purposes, an exponentially declining SFR with an e-folding timescale of 0.5 Gyr
28
Figure 8: A schematic illustration of the evolution of the stellar birthrate for different Hubble types.
The left panel shows the evolution of the relative SFR with time, following Sandage (1986). The
curves for spiral galaxies are exponentially declining SFRs that fit the mean values of the birthrate
parameter b measured by Kennicutt et al (1994). The curve for elliptical galaxies and bulges is
an arbitrary dependence for a e-folding time of 0.5 Gyr, for comparative purposes only. The right
panel shows the corresponding evolution in SFR with redshift, for an assumed cosmological density
parameter Ω = 0.3 and an assumed formation redshift zf = 5.
is also shown, as might be appropriate for an old spheroidal population. In this simple picture
the Hubble sequence is primarily one dictated by the characteristic timescale for star formation.
In the more contemporary hierarchical pictures of galaxy formation, these smooth histories would
be punctuated by merger-induced starbursts, but the basic long-term histories would be similar,
especially for the disks.
The righthand panel in Figure 8 shows the same star formation histories, but transformed into
SFRs as functions of redshift (assuming Ω=0.3 and a formation redshift zf =5). This diagram
illustrates how the dominant star forming (massive) host galaxy populations might evolve with
redshift. Most star formation at the present epoch resides in late-type gas-rich galaxies, but by
z ∼ 1 all spiral types are predicted to have comparable SFRs, and (present-day) early-type systems
become increasingly dominant at higher redshifts. The tendency of early-type galaxies to have
higher masses will make the change in population with redshift even stronger. It will interesting to
see whether these trends are observed in future lookback studies. Many readers are probably aware
that the redshift dependence of the volume averaged SFR shows quite a different character, with a
broad maximum between z∼1–2 and a decline at higher redshifts (Madau et al 1996, 1998). This
difference probably reflects the importance of hierarchical processes such as mergers in the evolution
of galaxies, mechanisms which are not included in the simple phenomenological description in Figure
8 (Pei & Fall 1995, Madau et al 1998).
29
5.2
Evolution of Circumnuclear Star Formation
The SFRs in the circumnuclear regions are largely decoupled from those of disks, and show no
strong relationship to either the gas contents or the bulge/disk properties of the parent galaxies.
Instead, the nuclear SFRs are closely associated with dynamical influences such as gas transport by
bars or external gravitational perturbations, which stimulate the flow of gas into the circumnuclear
regions.
The temporal properties of the star formation in the nuclear regions show a wide variation. Approximately 80–90% of spiral nuclei in optically-selected samples exhibit modest levels of Balmer
emission, with an average Hα emission-line equivalent width of 20–30 Å (Stauffer 1982, Kennicutt
et al 1989b, Ho et al 1997a, b). This is comparable to the average value in the disks of late-type
spiral galaxies, and is in the range expected for constant star formation over the age of the disk
(Kennicutt 1983a, Kennicutt et al 1994). Hence most nuclei show SFRs consistent with steadystate or declining star formation, though it is likely that some of these nuclei are observed in a
quiescent stage between major outbursts.
Starbursts are clearly the dominant mode of star formation in IR-selected samples of nuclei. The
typical gas consumption times are in the range 108 − 109 years (Figure 7), so the high SFRs can
only be sustained for a small percentage of the Hubble time. These timescales can be extended if
a steady supply is introduced from the outside, for example by a strong dissipative bar. The most
luminous nuclear starbursts (Lbol ≥1012 L⊙ ) are singular events. Maintaining such luminosities for
even 108 years requires a total gas mass on the order of 1010 –1011 M⊙ , equivalent to the total
gas gas supply in most galaxies. Violent interactions and mergers are the only events capable of
triggering such a catastrophic mass transfer.
5.3
Physical Regulation of the SFR
Although star forming galaxies span millionfold ranges in their present SFRs and physical conditions, there is a remarkable continuity in some of the their properties, and these relationships
provide important insights into the physical regulation of the SFR over this entire spectrum of
activities.
We have already seen evidence from Figures 5 and 7 that the global SFRs of disks and nuclear
starbursts are correlated with the local gas density, though over very different ranges in density
and SFR per unit area. The left panel of Figure 9 shows both sets of data plotted on a common
scale, and it reveals that the entire range of activities, spanning 5–6 orders of magnitude in gas
and SFR densities, fit on a common power law with index N ∼1.4 (Kennicutt 1998). The SFRs
for the two sets of data were derived using separate methods (Hα luminosities for the normal disks
and FIR luminosities for the starbursts), and to verify that they are measured on a self-consistent
scale, Figure 9 also shows Hα-derived SFRs gas densities for the centers (1–2 kpc) of the normal
disks (plotted as open circles). The tight relation shows that a simple Schmidt (1959) power law
provides an excellent empirical parametrization of the SFR, across an enormous range of SFRs,
and it suggests that the gas density is the primary determinant of the SFR on these scales.
30
Figure 9: (Left) The global Schmidt law in galaxies. Solid points denote the normal spirals in Figure
5, squares denote the circumnuclear starbursts in Figure 7. The open circles show the SFRs and
gas densities of the central regions of the normal disks. (Right) The same SFR data, but plotted
against the ratio of the gas density to the average orbital time in the disk. Both plots are adapted
from Kennicutt (1998).
The uncertainty in the slope of the best fitting Schmidt law is dominated by systematic errors in the
SFRs, with the largest being the FIR-derived SFRs and CO-derived gas densities in the starburst
galaxies. Changing either scale individually by a factor of two introduces a change of 0.1 in the
fitted value of N , and this is a reasonable estimate of the systematic errors involved (Kennicutt
1998). Incorporating these uncertainties yields the following relation for the best-fitting Schmidt
law:
Σgas
)1.4±0.15 M⊙ yr−1 kpc−2 .
(7)
ΣSF R = (2.5 ± 0.7) × 10−4 (
1 M⊙ pc−2
where ΣSF R and Σgas are the disk-averaged SFR and gas densities, respectively.
As discussed by Larson (1992) and Elmegreen (1994), a large-scale Schmidt law with index N ∼1.5
would be expected for self-gravitating disks, if the SFR scales as the ratio of the gas density (ρ)
to the free-fall timescale (∝ ρ−0.5 ), and the average gas scale height is roughly constant across the
sample (Σ ∝ ρ). In a variant on this picture, Elmegreen (1997) and Silk (1997) have suggested
that the SFR might scale with the ratio of the gas density to the average orbital timescale; this
is equivalent to postulating that disks process a fixed fraction of their gas into stars in each orbit
around the galactic center. The right panel of Figure 9, also adapted from Kennicutt (1998), shows
the correlation between the SFR density and Σgas /τdyn for the same galaxies and starbursts. For
this purpose τdyn is defined as one disk orbit time, measured at half of the outer radius of the
star forming disk, in units of years (see Kennicutt 1998 for details). The line is a median fit to the
31
normal disks with slope contrained to unity, as predicted by the simple Silk model. This alternative
“recipe” for the SFR provides a fit that is nearly as good as the Schmidt law. The equation of the
fit is:
ΣSF R = 0.017 Σg Ωg .
(8)
In this parametrization the SFR is simply ∼10% of the available gas mass per orbit.
These parametrizations offer two distinct interpretations of the high SFRs in the centers of luminous
starburst galaxies. In the context of the Schmidt law picture, the star formation efficiency scales
(N −1)
as Σg
, or Σ0.4
g for the observed relation in Figure 9. The central starbursts have densities that
are on the order of 100–10000 times higher than in the extended star forming disks of spirals, so
the global star formation efficiencies should be 6–40 times higher. Alternatively, in the kinematical
picture, the higher efficiencies in the circumnuclear starbursts are simply a consequence of the
shorter orbital timescales in the galaxy centers, independent of the gas density. Whether the
observed Schmidt law is a consequence of the shorter dynamical times or vice versa cannot be
ascertained from these data alone, but either description provides an excellent empirical description
or “recipe” for the observed SFRs.
These simple pictures can account for the high SFRs in the starburst galaxies, as well as for the
observed radial variation of SFRs within star forming disks (Kennicutt 1989, 1997). However
the relatively shallow N ∼1.4 Schmidt law cannot account for the strong changes in disk SFRs
observed across the Hubble sequence, if the disks evolved as nearly closed systems (Kennicutt et
al 1994). Likewise the modest changes in galaxy rotation curves with Hubble type are too small
to account for the large differences in star formation histories with a kinematical model such as in
equation [8]. The explanation probably involves star formation thresholds in the gas-poor galaxies
(Kennicutt 1989, Kennicutt et al 1997), but the scenario has not been tested quantitatively, and it
is possible that other mechanisms, such as infall of gas, merger history, or bulge-disk interactions
are responsible for the strong changes in star formation histories across the spiral sequence.
6
FUTURE PROSPECTS
The observations described in this review have provided us with the beginnings of a quantitative picture of the star formation properties and evolutionary properties of the Hubble sequence.
However the picture remains primitive in many respects, being based in large part on integrated,
one-zone averages over entire galaxies, and extrapolations from present-day SFRs to crude characterizations of the past star formation histories. Uncertainties in fundamental parameters such as
the IMF and massive stellar evolution undermine the accuracy of the entire SFR scale, and weaken
the interpretations that are based on these measurements. Ongoing work on several fronts should
lead to dramatic progress over the next decade, however.
The most exciting current development is the application of the SFR diagnostics described in Section 2 to galaxies spanning the full range of redshifts and lookback times (Ellis 1997). This work has
already provided the first crude measures of the evolution in the volume-averaged SFR (Madau et al
32
1996, 1998). The combination of 8–10 meter class groundbased telescopes, HST, and eventually the
Next Generation Space Telescope should eventually provide detailed inventories of integrated spectra, SFRs and morphologies for complete samples of galaxies at successive redshifts. This should
give the definitive picture of the star formation history of the Hubble sequence, and impose strong
tests on galaxy formation and evolution models. At the same time, a new generation of IR space
observatories, including the Wide Field Infrared Explorer and the Space Infrared Telsescope Facility, will provide high-resolution observations of nearby starburst galaxies, and the first definitive
measurements of the cosmological evolution of the infrared-luminous starburst galaxy population.
Although studies of the star formation histories of nearby galaxies are largely being supplanted by
the more powerful lookback studies, observations of nearby galaxies will remain crucial for understanding many critical aspects of galaxy formation and evolution. Perhaps the greatest potential
is for understanding the physical processes that determine the local and global SFRs in galaxies,
and understanding the feedback processes between the star formation and the parent galaxies.
This requires spatially-resolved measurements of SFRs over the full spectrum of insterstellar and
star formation environments, and complementary measurements of the densities, dynamics, and
abundances of the interstellar gas. Uncertainty about the nature of the star formation law and the
SFR–ISM feedback cycle remain major stumbling blocks to realistic galaxy evolution models, but
observations over the next decade should provide the foundations of a physically-based model of
galactic star formation and the Hubble sequence.
7
ACKNOWLEDGEMENTS
I wish to express special thanks to my collaborators in the research presented here, especially my
current and former graduate students Audra Baleisis, Fabio Bresolin, Charles Congdon, Murray
Dixson, Kevin Edgar, Paul Harding, Crystal Martin, Sally Oey, Anne Turner, and Rene Walterbos. During the preparation of this review my research was supported by the National Science
Foundation though grant AST-9421145.
33
Literature Cited
Aalto S, Booth RS, Black JH, Koribalski B., Wielebinski R. 1994. Astron. Astrohys. 286:365-80
Athanassoula E. 1992. MNRAS 259:345-64
Bagnuolo WG. 1976. The Stellar Content and Evolution of Irregular and Other Late-Type Galaxies.
PhD thesis. Caltech
Balzano VA. 1983. Ap. J. 268:602-27
Barnes JE, Hernquist L. 1992. Annu. Rev. Astron. Astrophys. 30:705-42
Bechtold J, Yee HKC, Elston R, Ellingson E. 1997. Ap. J. Lett. 477:L29-L32
Bertelli G, Bressan A, Chiosi C, Fagotto F, Nasi E. 1994. Astron. Astrophys. Suppl. 106:275-302
Boselli A. 1994. Astron. Astrophys. 292:1-12
Boselli A, Gavazzi G, Lequeux J, Buat V, Casoli F, et al. 1995. Astron. Astrophys. 300:L13-L16
Bothun GD. 1990. In Evolution of the Universe of Galaxies, ed. RG Kron, ASP Conf. Proc.
10:54-66, San Francisco: Astron. Soc. Pac.
Bresolin F, Kennicutt RC. 1997, Astron. J. 113:975-80
Bruzual G, Charlot S. 1993. Ap. J. 405:538-53
Buat V. 1992. Astron. Astrophys. 264:444-54
Buat V, Deharveng JM. 1988. Astron. Astrophys. 195:60-70
Buat V, Deharveng JM, Donas J. 1989. Astron. Astrophys. 223:42-46
Buat V, Xu C. 1996. Astron. Astrophys. 306:61-72
Bushouse HA. 1986. Astron. J. 91:255-70
Bushouse HA. 1987. Ap. J. 320:49-72
Bushouse HA, Werner MW, Lamb SA. 1988. Ap. J. 335:74-92
Caldwell N, Kennicutt R, Phillips AC, Schommer RA. 1991. Ap. J. 370:526-40
Caldwell N, Kennicutt R, Schommer R. 1994. Astron.J. 108:1186-90
Calzetti D, Kinney AL, Storchi-Bergmann T. 1994. Ap.J. 429:582-601
Calzetti D, Kinney AL, Storchi-Bergmann T. 1996. Ap. J. 458:132-5
Calzetti D. 1997. Astron. J. 113:162-84
Caplan J, Deharveng L. 1986. Astron. Astrophys. 155:297-313
Caplan J, Ye T, Deharveng L, Turtle AJ, Kennicutt RC. 1996. Astron. Astrophys. 307:403-16
Carico DP, Sanders DB, Soifer BT, Matthews K, Neugebauer G. 1990. Astron. J. 100:70-83
Cayatte V, Kotanyi C, Balkowski C, van Gorkom JH. 1994. Astron. J. 107:1003-17
Cepa J, Beckman JE 1990. Ap. J. 349:497-502
Charlot S, Bruzual G. 1991. Ap. J. 367:126-40
34
Cohen JG. 1976. Ap. J. 203:587-92
Cox, P, Mezger PG. 1989. Astron. Astrophys. Rev. 1:49-83
Cowie LL, Hu EM, Songaila A, Egami E. 1997. Ap. J. Lett. 481:L9-L13
Cowie LL, Songaila A, Hu EM, Cohen JG. 1996. Astron. J. 112:839-64
Cutri RM, McAlary CW. 1985. Ap. J. 296:90-105
Deharveng JM, Sasseen TP, Buat V, Bowyer S, Lampton M, Wu X. 1994. Astron. Astrophys.
289:715-728
de Vaucouleurs G, de Vaucouleurs A, Corwin HG. 1976. Second Reference Catalog of Bright Galaxies. Austin: Univ. of Texas Press (RC2)
Devereux N. 1987. Ap. J. 323:91-107
Devereux NA, Becklin EE, Scoville N. 1987. Ap. J. 312:529-41
Devereux NA, Hameed S. 1997. Astron. J. 113:599-608
Devereux NA, Young JS. 1990. Ap. J. Lett. 350:L25-28
Devereux NA, Young JS. 1991. Ap. J. 371:515-24
Donas J, Deharveng JM. 1984. Astron. Astrophys. 140:325-333
Donas J, Deharveng JM, Laget M, Milliard B, Huguenin D. 1987. Astron. Astrophys. 180:12-26
Donas J, Milliard B, Laget M, Buat V. 1990. Astron. Astrophys. 235:60-68
Donas J, Milliard B, Laget M. 1995. Astron. Astrophys. 303:661-672
Downes D, Solomon PM, Radford SJE. 1993. Ap. J. Lett. 414:L13-L16
Dressel LL. 1988. Ap. J. Lett. 329:L69-L73
Dressler A. 1984. Annu. Rev. Astron. Astrophys. 22:185-222
Ellis RS. 1997. Annu. Rev. Astron. Astrophys. 35:389-443
Elmegreen BG. 1994. Ap. J. Lett. 425:L73-76
Elmegreen BG. 1997, In Starburst Activity in Galaxies, ed J Franco, R Terlevich, A Serrano, Rev.
Mex. Astron. Astrophys. Conf. Ser. 6:165-71
Elmegreen BG, Elmegreen DM. 1986. Ap. J. 311:554-562
Engelbracht CW. 1997. Infrared Observations and Stellar Populations Modelling of Starburst
Galaxies. PhD thesis, Univ. Arizona
Evans IN, Koratkar AP, Storchi-Bergmann T, Kirkpatrick H, Heckman TM, Wilson AS. 1996. Ap.
J. Suppl. 105:93-127
Fanelli MN, Marcum PM, Waller WH, Cornett RH, O’Connell RW, et al. 1997. In The Ultraviolet
Universe at Low and High Redshift, ed. W Waller, M Fanelli, J Hollis, A Danks. New York: Am.
Inst. Phys.
Feinstein C. 1997. Ap. J. Suppl. 112:29-47
35
Ferguson AMN, Wyse RFG, Gallagher JS, Hunter DA. 1996. Astron. J. 111:2265-79
Fioc M, Rocca-Volmerange B. 1997. Astron. Astrophys. 326:950-62
Friedli D, Benz W. 1995. Astron. Astrophys. 301:649-65
Gallagher JS, Hunter DA. 1984. Ann. Rev. Astron. Astrophys. 22:37-74
Gallagher JS, Hunter DA, Bushouse H. 1989. Astron. J. 97:700-07
Gallagher JS, Hunter DA, Tutukov AV. 1984. Ap. J. 284:544-56
Gallego J, Zamorano J, Aragon-Salamanca A, Rego M. 1995. Ap. J. Lett. 445:L1-L4
Gallimore JF, Keel WC. 1993. Astron. J. 106:1337-43
Gavazzi G, Boselli A, Kennicutt R. 1991. Astron. J. 101:1207-30
Gavazzi G, Jaffe W. 1985. Ap. J. Lett. 294:L89-L92
Gavazzi G, Pierini D, Boselli A. 1996. Astron. Astrophys. 312:397-408
Gavazzi G, Scodeggio M. 1996. Astron. Astrophys. 312:L29-L32
Gilmore GF, Howell DJ. (eds.) 1998. The Stellar Initial Mass Function., ASP Conf. Proc., Vol.
142. San Francisco: Astron. Soc. Pac.
Giuricin G, Tamburini L, Mardirossian F, Mezzetti M, Monaco P. 1994. Astron. Astrophys.
427:202-20
Goldader JD, Joseph RD, Doyon R, Sanders DB. 1995. Ap. J. 444:97-112
Goldader JD, Joseph RD, Doyon R, Sanders DB. 1997. Ap. J. Suppl. 108:449-70
González Delgado, RM, Perez E, Tadhunter C, Vilchez J, Rodríguez-Espinoza JM. 1997. Ap. J.
Suppl. 108:155-98
Hawarden TG, Mountain CM, Leggett SK, Puxley PJ. 1986. MNRAS 221:41P-45P
Harper DA, Low FJ. 1973. Ap. J. Lett. 182:L89-L93
Haynes MP, Giovanelli R, Chincarini GL. 1984. Ann. Rev. Astron. Astrophys. 22:445-70
Heckman TM 1990. In Paired and Interacting Galaxies, IAU Colloq. 124, ed. JW Sulentic, WC
Keel, CM Telesco, NASA Conf. Publ. CP-3098, pp. 359-82. Washington DC: NASA
Heckman TM. 1994. In Mass-Transfer Induced Activity in Galaxies, ed. I Shlosman, pp. 234-50.
Cambridge: Cambridge Univ. Press
Heckman TM, Crane PC, Balick B. 1980. Astron. Astrophys. Suppl. 40:295-305
Heiles C. 1990. Ap. J. 354:483-91
Ho LC, Filippenko AV, Sargent WLW. 1997. Ap. J. 487:579-90
Ho LC, Filippenko AV, Sargent WLW. 1997. Ap. J. 487:591-602
Ho PTP, Beck SC, Turner JL. 1990. Ap. J. 349:57-66
Hodge PW. 1989. Annu. Rev. Astron. Astrophys. 27:139-59
Hodge PW, Kennicutt RC. 1983. Astron. J. 88:296-328
36
Huang JH, Gu QS, Su HJ, Hawarden TG, Liao XH, Wu GX. 1996: Astron. Astrophys. 313:13-24
Hubble E. 1926. Ap. J. 64:321-69
Huchra JP. 1977. Ap. J. 217:928-39
Hunter DA. 1994. Astron. J. 107:565-81
Hunter DA, Gallagher JS. 1985. Ap. J. Suppl. 58:533-60
Hunter DA, Gallagher JS. 1990. Ap. J. 362:480-90
Hunter DA, Gillett FC, Gallagher JS, Rice WL, Low FJ. 1986. Ap. J. 303:171-85
Hunter DA, Hawley WN, Gallagher JS. 1993. Astron. J. 106:1797-1811
Isobe T, Feigelson E. 1992. Ap. J. Suppl. 79:197-211
Israel FP, van der Hulst JM. 1983. Astron. J. 88:1736-48
Joseph RD, Wright GS. 1985. MNRAS 214:87-95
Kaufman M, Bash FN, Kennicutt RC, Hodge PW. 1987. Ap. J. 319:61-75
Keel WC. 1983. Ap. J. 269:466-86
Keel WC, Kennicutt RC, Hummel E, van der Hulst JM. 1985. Astron. J. 90:708-30
Kennicutt RC. 1983a. Ap. J. 272:54-67
Kennicutt RC. 1983b. Astron. J. 88:483-88
Kennicutt RC. 1989. Ap. J. 344:685-703
Kennicutt RC. 1992a. Ap. J. 388:310-27
Kennicutt RC. 1992b. Ap. J. Suppl. 79:255-84
Kennicutt RC. 1997, In The Interstellar Medium in Galaxies, ed. JM van der Hulst, pp. 171-95.
Dordrecht: Kluwer
Kennicutt RC. 1998. Ap. J. 498:541-52
Kennicutt RC, Schweizer F, Barnes JE. 1998. Galaxies: Interactions and Induced Star Formation,
Saas-Fee Advanced Course 26, ed. D Friedli, L Martinet, D Pfenniger, Berlin:Springer
Kennicutt RC, Bothun GD, Schommer RA. 1984. Astron. J. 89:1279-87
Kennicutt RC, Bresolin F, Bomans DJ, Bothun GD, Thompson IB. 1995. Astron. J. 109:594-604
Kennicutt RC, Edgar BK, Hodge PW. 1989a. Ap. J. 337:761-81
Kennicutt RC, Keel WC, Blaha CA. 1989b. Astron. J. 97:1022-35
Kennicutt RC, Keel WC, van der Hulst JM, Hummel E, Roettiger KA. 1987. Astron. J. 93:1011-23
Kennicutt RC, Kent SM. 1983. Astron. J. 88:1094-1107
Kennicutt RC, Tamblyn P, Congdon CW. 1994. Ap. J. 435:22-36
Kinney AL, Bohlin RC, Calzetti D, Panagia N, Wyse RFG. 1993. Ap. J. Suppl. 86:5-93
Klein U, Grave R. 1986. Astron. Astrophys. 161:155-68
37
Knapen J, Beckman JE, Cepa J, van der Hulst JM, Rand RJ. 1992. Ap. J. Lett. 385:L37-L40
Larson RB. 1992, In Star Formation in Stellar Systems, ed. G Tenorio-Tagle, M Prieto, F Sánchez.
pp. 125-190. Cambridge: Cambridge Univ. Press
Larson RB, Tinsley BM. 1978. Ap. J. 219:46-59
Lawrence A, Rowan-Robinson M, Leech K, Jones DHP, Wall JV. 1989. MNRAS 240:329-48
Leech M, Rowan-Robinson M, Lawrence A, Hughes JD. 1994. MNRAS 267:253-69
Lehnert MD, Heckman TM. 1996. Ap. J. 472:546-63
Leitherer C, Ferguson HC, Heckman TM, Lowenthal JD. 1995a. Ap. J. Lett. 454:L19-L22
Leitherer C, Fritze-v. Alvensleben U, Huchra JP. (eds.) 1996b. From Stars to Galaxies: The
Impact of Stellar Physics on Galaxy Evolution. ASP Conf. Proc. Vol. 98. San Francisco: Astron.
Soc Pac.
Leitherer C, Heckman TM. 1995. Ap. J. Suppl. 96:9-38
Leitherer C, Robert C, Heckman TM. 1995b. Ap. J. Suppl. 99:173-87
Leitherer C, Alloin D, Alvensleben UF, Gallagher JS, Huchra JP, et al. 1996a. Publ. Astron. Soc.
Pac. 108:996-1017
Liu CT, Kennicutt RC. 1995. Ap. J. 450:547-58
Lonsdale CJ, Helou G. 1987. Ap. J. 314:513-24
Lonsdale CJ, Persson SE, Matthews K. 1984. Ap. J. 287:95-107
Lutz D, Genzel R, Sternberg A, Netzer H, Kunze D, et al. 1996. Astron. Astrophys. 315:L137-L140
Madau P, Ferguson H, Dickinson M, Giavalisco M, Steidel CC, Fruchter A. 1996. MNRAS 283:13881404
Madau P, Pozzetti L, Dickinson M. 1998. Ap. J. in press
Maoz D, Filippenko AV, Ho LC, Macchetto D, Rix H-W, Schneider DP. 1996. Ap. J. Suppl.
107:215-26
Martin CL. 1997. Ap. J. in press
Martinet L, Friedli D. 1997. Astron. Astrophys. 323:363-73
Massey P. 1998. In The Stellar Initial Mass Function. ed. BF Gilmore, DJ Howell. ASP Conf.
Proc. 142:17-44. San Francisco: Astron. Soc. Pac.
McCall ML, Schmidt FH. 1986. Ap. J. 311:548-53
Meurer GR, Heckman TM, Leitherer C, Kinney A, Robert C, Garnett DR. 1995. Astron. J.
110:2665-91
Meurer GR, Gerhardt R, Heckman TM, Lehnert MD, Leitherer C, Lowenthal J. 1997, Astron. J.
114:54-68
Mihos JC, Hernquist L. 1996. Ap. J. 464:641-63
Morgan WW. 1958. Publ. Astron. Soc. Pac. 70:364-91
38
Moss C, Whittle M. 1993. Ap. J. Lett.200 407:L17-L20
Moshir M, Kopan, G, Conrow J, McCallon H, Hacking P, et al. 1992. Explanatory Supplement to
the IRAS Faint Source Survey, Version 2, JPL D-10015 8/92, (Pasadena: JPL)
Niklas S, Klein U, Braine J, Wielebinski R. 1995. Astron. Astrophys. Suppl. 114:21-49
Niklas S, Klein U, Wielebinski R. 1997. Astron. Astrophys. 322:19-28
Norman C, Ikeuchi S. 1989. Ap. J. 345:372-83
Oey MS, Kennicutt RC. 1997. MNRAS 291:827-32
Ostriker JP, Thuan TX. 1975. Ap. J. 202:353-64
Patel K, Wilson CD. 1995a. Ap. J. 451:607-15
Patel K, Wilson CD. 1995b. Ap. J. 453:162-72
Pei YC, Fall SM. 1995. Ap. J. 454:69-76
Phillips AC. 1993. Star Formation in Barred Spiral Galaxies. PhD thesis, Univ. Washington,
Seattle
Pogge RW, Eskridge PB. 1987. Astron. J. 93:291-300
Pogge RW, Eskridge PB. 1993. Astron. J. 106:1405-19
Pompea SM, Rieke GH. 1990. Ap. J. 356:416-29
Puxley PJ, Brand PWJL, Moore TJT, Mountain CM, Nakai N, Yamashita AT. 1989. Ap. J.
345:163-68
Puxley PJ, Hawarden TG, Mountain CM. 1990. Ap. J. 364:77-86
Rieke GH, Lebofsky MJ. 1978. Ap. J. Lett. 220:L37-L41
Rieke GH, Lebofsky MJ. 1979. Ann. Rev. Astron. Astrophys. 17:477-511
Rieke, GH, Loken K, Rieke MJ, Tamblyn P. 1993. Ap. J. 412:99-110
Rieke GH, Low FJ. 1972. Ap. J. Lett. 176:L95-L100
Roberts MS. 1963. Ann. Rev. Astron. Astrophys. 1:149-78
Roberts MS, Haynes MP. 1994. Ann. Rev. Astron. Astrophys. 32:115-52
Romanishin W. 1990. Astron. J. 100:373-76
Rowan-Robinson M, Crawford J. 1989. MNRAS 238:523-58
Rubin VC, Kenney JDP, Young JS. 1997. Astron. J. 113:1250-78
Ryder SD. 1993. Massive Star Formation in Galactic Disks. PhD thesis. Australian National
Univ.
Ryder SD, Dopita MA. 1993. Ap. J. Suppl. 88:415-21
Ryder SD, Dopita MA. 1994. Ap. J. 430:142-62
Salpeter EE. 1955. Ap. J. 121:161-67
Sandage A. 1986. Astron. Astrophys. 161:89-101
39
Sanders DB, Mirabel IF. 1996. Ann. Rev. Astron. Astrophys. 34:749-92
Sanders DB, Scoville NZ, Soifer BT. 1991. Ap. J. 370:158-71
Sanders DB, Soifer BT, Elias JH, Madore, BF, Matthews K, et al. 1988. Ap. J. 325:74-91
Sauvage M, Thuan TX. 1992. Ap. J. Lett. 396:L69-L73
Sauvage M, Thuan TX. 1994. Ap. J. 429:153-71
Scalo JM. 1986. Fund. Cos. Phys. 11:1-278
Schmidt M. 1959. Ap. J. 129:243-58
Scoville NZ, Becklin EE, Young JS, Capps RW. 1983. Ap. J. 271:512-23
Scoville NZ, Hibbard JE, Yun MS, van Gorkom JH. 1994. In Mass-Transfer Induced Activity in
Galaxies, ed. I Shlosman, pp. 191-212. Cambridge: Cambridge Univ. Press
Searle L, Sargent WLW, Bagnuolo WG. 1973. Ap. J. 179:427-38
Sérsic JL, Pastoriza M. 1967. Publ. Astron. Soc. Pac 79:152-55
Silk J. 1997. Ap. J. 481:703-09
Smith AM, Cornett, RH. 1982, Ap. J. 261:1-11
Smith BJ, Harvey PM. 1996. Ap. J. 468:139-66
Smith EP, Pica AJ, Bohlin RC, Cornett RH, Fanelli MN. 1996. Ap. J. Suppl. 104:207-315
Soifer BT, Houck JR, Neugebauer G. 1987. Ann. Rev. Astron. Astrophys. 25:187-230
Solomon PM, Downes D, Radford SJE, Barrett JW. 1997. Ap. J. 478:144-61
Solomon PM, Sage LJ. 1988. Ap. J. 334:613-25
Stauffer JR. 1982. Ap. J. Suppl. 50:517-27
Steidel CC, Giavalisco M, Pettini M, Dickinson M, Adelberger KL. 1996. Ap. J. Lett. 462:L17-L21
Telesco CM. 1988. Ann. Rev. Astron. Astrophys. 26:343-76
Telesco CM, Dressel LL, Wolstencroft RD. 1993. Ap. J. 414:120-43
Telesco CM, Harper DA. 1980. Ap. J. 235:392-404
Telesco CM, Wolstencroft RD, Done C. 1988. Ap. J. 329:174-86
Thronson HA, Bally J, Hacking P. 1989. Astron. J. 97:363-74
Tinney CG, Scoville NZ, Sanders DB, Soifer BT 1990, Ap. J. 362:473-79
Tinsley BM. 1968. Ap. J. 151:547-65
Tinsley BM. 1972. Astron. Astrophys. 20:383-96
Tomita A, Tomita Y, Saito M. 1996. Pub. Astron. Soc. J. 48:285-303
Tully RB, Mould JR, Aaronson M. 1982. Ap. J. 257:527-37
Turner JL, Ho PTP. 1994. Ap. J. 421:122-39
van der Hulst JM, Kennicutt RC, Crane PC, Rots AH. 1988. Astron. Astrophys. 195:38-52
40
Veilleux S, Kim D-C, Sanders DB, Mazzarella JM, Soifer BT. 1995. Ap. J. Suppl. 98:171-217
Waller W, Fanelli M, Danks A, Hollis J. 1997. The Ultraviolet Universe at Low and High Redshift,
AIP Conf. 408. New York: Am. Inst. Phys.
Walterbos RAM, Braun R. 1994. Ap. J. 431:156-71
Walterbos RAM, Greenawalt B. 1996. Ap. J. 460:696-710
Warmels RH. 1988. Astron. Astrophys. Suppl. 72:427-47
Weedman DW, Feldman FR, Balzano VA, Ramsey LW, Sramek RA, Wu C-C. 1981. Ap. J.
248:105-12
Whitford AE. 1975. in Galaxies in the Universe, ed. A Sandage, M Sandage, J Kristian, Stars
Stellar Syst. Compend. 9:159-76. Chicago: Univ. Chicago Press
Wright GS, Joseph RD, Robertson NA, James PA, Meikle WPS. 1988. MNRAS 233:1-23
Wyse RFG. 1983. MNRAS 199:1P-8P
Xu C, Sulentic JW. 1991. Ap. J. 374:407-30
Young JS, Allen L, Kenney JDP, Lesser A, Rownd B. 1996. Astron. J. 112:1903-27
Young JS, Scoville NZ. 1991. Ann. Rev. Astron. Astrophys. 29:581-625
Young JS, Schloerb, FP, Kenney JDP, Lord SD. 1986. Ap. J. 304:443-58
41